Three publishing houses, Cambridge University Press, Oxford University Press, and Sage Publications, Inc. (collectively, “Plaintiffs”) allege that members of the Board of Regents of the University System of Georgia and officials at Georgia State University (“GSU”) (collectively, “Defendants”) infringed Plaintiffs’ copyrights by maintaining a policy which allows GSU professors to make digital copies of excerpts of Plaintiffs’ books available to students without paying Plaintiffs. Plaintiffs alleged seventy-four individual instances of infringement, which took place during three academic terms in 2009. The District Court issued an order finding that Plaintiffs failed to establish a prima facie case of infringement in twenty-six instances, that the fair use defense applied in forty-three instances, and that Defendants had infringed Plaintiffs’ copyrights in the remaining five instances.
Finding that GSU’s policy caused the five instances of infringement, the District Court granted declaratory and injunctive relief to Plaintiffs. Nevertheless, the District Court found that Defendants were the prevailing party and awarded them costs and attorneys’ fees. Because we find that the District Court’s fair use analysis was in part erroneous, we reverse the District Court’s judgment; vacate the injunction, declaratory relief, and award of costs and fees; and remand for further proceedings consistent with this opinion.
I.
A.
Like many recent issues in copyright law, this is a case in which technological advances have created a new, more efficient means of delivery for copyrighted works, causing copyright owners and consumers to struggle to define the appropriate boundaries of copyright protection in the new digital marketplace. These boundaries must be drawn carefully in order to assure that copyright law serves its intended purpose, which is to promote the creation of new works for the public good by providing authors and other creators with an economic incentive to create. See Twentieth Century Music Corp. v. Aiken, 422 U.S. 151, 156, 95 S.Ct. 2040, 2044, 45 *1238L.Ed.2d 84 (1975). If copyright’s utilitarian goal is to be met, we must be careful not to place overbroad restrictions on the use of copyrighted works, because to do so would prevent would-be authors from effectively building on the ideas of others. Some unpaid use of copyrighted materials must be allowed in order to prevent copyright from functioning as a straightjacket that stifles the very creative activity it seeks to foster. If we allow too much unpaid copying, however, we risk extinguishing the economic incentive to create that copyright is intended to provide.
The fair use doctrine provides a means by which a court may ascertain the appropriate balance in a given case if the market actors cannot do so on their own. Fair use is a defense that can excuse what would otherwise be an infringing use of copyrighted material. See 17 U.S.C. § 107 (“[T]he fair use of a copyrighted work ... is not an infringement of copyright.”). To prevail on a claim of fair use, a defendant must convince the court that allowing his or her unpaid use of copyrighted material would be equitable and consonant with the purposes of copyright. In order to make this determination, the court must carefully evaluate the facts of the case at hand in light of four considerations, which are codified in the Copyright Act of 1976: (1) the purpose of the allegedly infringing use, (2) the nature of the original work, (3) the size and significance of the portion of the original work that was copied, and (4) the effect of the allegedly infringing use on the potential market for or value of the original. Id. These factors establish the contours within which a court may investigate whether, in a given case, a finding of fair use would serve the objectives of copyright. Here, we are called upon to determine whether the unpaid copying of scholarly works by a university for use by students — facilitated by the development of systems for digital delivery over the Internet — should be excused under the doctrine of fair use.
Plaintiffs are three publishing houses that specialize in academic works. Plaintiff Cambridge University Press (“Cambridge”) is the not-for-profit publishing house of the University of Cambridge in England, having an American branch headquartered in New York City. Plaintiff Oxford University Press, Inc. (“Oxford”) is a not-for-profit United States corporation associated with Oxford University in England and headquartered in New York City. Plaintiff Sage Publications, Inc. (“Sage”) is a for-profit Delaware corporation, headquartered in Sherman Oaks, California.
Plaintiffs do not publish the large, general textbooks commonly used in entry-level university courses. Rather, Plaintiffs publish advanced scholarly works, which might be used in upper-level undergraduate and graduate courses. Cambridge and Oxford publish scholarly books and journals on niche subject areas. Their works involved in this case include research-based monographs, which aré “small, single author books which give in-depth analysis of a narrow topic,” Cambridge Univ. Press v. Becker, 863 F.Supp.2d 1190, 1212 (N.D.Ga.2012) (footnote omitted), instructional books, trade books, and other works on academic topics. Sage primarily publishes books on the social sciences. AH' three plaintiffs publish, in addition to works by a single author, “edited books” which feature the contributions of multiple authors. Id.
Plaintiffs market their books to professors who teach at universities and colleges. Cambridge and Oxford regularly send complimentary copies of their publications to professors. Sage provides trial copies upon request. Plaintiffs intend that professors use Plaintiffs’ publications in their *1239work ánd assign them as required reading so that students will purchase them.
Rather than assigning whole books, some professors assign or suggest excerpts from Plaintiffs’ books as part of the curriculum for their courses. Professors might do this by putting the work on reserve at the university library so that students can visit the library to read an assigned excerpt. Or, professors might prepare a bound, photocopied, paper “coursepack” containing excerpts from several works for a particular course. Often, a third-party copy shop assembles these coursepacks, performing the copying and binding, obtaining the necessary licenses from publishers, and charging students a fee for the finished coursepack. In recent years, however, universities— following the trend with regard to distribution of many forms of media the world over — have increasingly abandoned paper coursepacks in favor of digital distribution of excerpts over the Internet.1
GSU is a public university in Atlanta, Georgia. It is part of the University System of Georgia, and is overseen by the Board of Regents of the University System of Georgia. GSU maintains two on-campus systems known as “ERes” and “uLearn” for digital distribution of course materials to students.
ERes (short for “E-Reserves”) is an “electronic reserve system” hosted on servers maintained by GSU and managed by GSU’s library staff. Since 2004, GSU has used ERes to allow GSU students to access course materials — including course syllabi, class notes, sample exams, and excerpts from books and journals — on the Internet via a web browser. In order to place an excerpt from a book or journal on ERes, a professor must either provide a personal copy of the work to the GSU library staff or indicate that the GSU library owns a copy. A member of the library staff then scans the excerpt to convert it to a digital format and posts the scanned copy to ERes. GSU students are given access to an ERes website specific to the courses in which they are enrolled. On each course-specific ERes website, students find their reading assignments listed by title. The scanned excerpts are accessible via hyperlink. When a student clicks a link for a particular assignment, the student receives a digital copy of a scanned excerpt that the student may view, print, save to his or her computer, and potentially keep indefinitely. ERes course websites are password-protected in order to limit access to the students in the particular course. Once a course ends, students no longer have access to the ERes website for that course.
uLearn is a “course management system” hosted on servers maintained by the Board of Regents. Like ERes, uLearn provides course-specific webpages through which professors may make course material available, including digital copies of excerpts from books, which students in the course may view, print, or save. The most significant difference between the ERes and uLearn systems is that uLearn allows professors to upload digital copies of reading material directly to their course websites while ERes forces professors to rely *1240on GSU library personnel to upload reading material for them.
ERes and uLearn have been popular at GSU.2 For example, during the Spring 2009 term, paper coursepacks were offered for only about fifteen courses, while instructors in hundreds of courses made readings available on ERes. Thus, the excerpts from larger works that make up some portion of course readings at GSU, and which were once distributed to students via a paper coursepack purchased at the university bookstore, are now largely distributed to students via digital download on the Internet, that the students pay for only indirectly via tuition and fees.
There exists a well-established system for the licensing of excerpts of copyrighted works. Copyright Clearance Center (“CCC”) is ' a not-for-profit corporation with headquarters in Danvers, Massachusetts. CCC licenses excerpts from copyrighted works for a fee, acting on behalf of publishers who choose to make their works available through CCC. These licenses are called “permissions.” All three Plaintiffs offer excerpt-specific permissions to photocopy or digitally reproduce portions of their works, which may be obtained directly from Plaintiffs or through CCC. Permissions are not, however, available for licensed copying of excerpts from all of Plaintiffs’ works.3
CCC offers a variety of permissions services to various categories of users, including corporate, educational, and institutional users. One such service, the Academic Permissions Service (“APS”), licenses educational users to make print copies on a per-use basis. CCC also offers an electronic course content service (“ECCS”) for licensing of digital excerpts by educational users on a per-use basis, that — in 2008, the year for which evidence on the question was presented — offered only a small percentage of the works that were available through APS. ECCS is designed for electronic reserve systems such as ERes and uLearn. Software is available that would allow GSU library personnel to place an order with CCC for a permission to provide students with a digital copy of an excerpt via ERes. CCC also offers an Academic Repertory License Service (“ARLS”) which affords subscribers access to excerpts from a set group of about nine million titles, approximately 17 percent of which are available in digital format. Sage *1241participates in ARLS and did so in 2009, Oxford participated in 2009 with regard to journals but not books, and Cambridge does not participate. GSU did not and does not subscribe to this program.
When the GSU bookstore assembles and sells a paper coursepack containing excerpts from copyrighted works, GSU pays permissions fees for use of the excerpts.4 The central issue in this case is under what circumstances GSU must pay permissions fees to post a digital copy of an excerpt of Plaintiffs’ works to ERes or uLearn.
B.
On April 15, 2008, Plaintiffs filed their original complaint in the United States District Court for the Northern District of Georgia. Plaintiffs alleged that hundreds of GSU professors have made thousands of copyrighted works — including works owned or controlled by Plaintiffs— available on GSU’s electronic reserve systems without obtaining permissions from copyright holders, and- that GSU’s administration facilitated, encouraged, and induced this practice.' Plaintiffs sued Defendants in their official capacities as GSU officials, claiming (1) direct copyright infringement 5 caused by the officials “scanning,- copying, displaying, and distributing Plaintiffs’ copyrighted material;” (2) contributory copyright infringement6 caused by the officials’ “facilitating, encouraging, and inducing librarians and professors to *1242scan, copy, display, and distribute Plaintiffs’ copyrighted material” and “students to view, download, copy and further distribute [Plaintiffs’] copyrighted material;” and (8) vicarious copyright infringement7 caused by the officials’ inducing GSU employees and students to copy Plaintiffs’ copyrighted material, profiting from this practice, and failing to stop it despite having the right and ability to do so. Doc. 1, at 25-28. Plaintiffs sought declaratory and injunctive relief. ’ Plaintiffs supported their allegations with numerous examples of GSU professors posting excerpts of Plaintiffs’ works on GSU’s electronic reserve systems. Defendants filed an Answer, denying Plaintiffs’ allegations of infringement; claiming sovereign immunity and Eleventh Amendment immunity based on Defendants’ status as state officials; and asserting a defense of fair use because “any alleged use of copyrighted materials was for the purpose of teaching, scholarship or research and for nonprofit educational purposes.” Doc. 14, at 2.
On December 15, 2008, Plaintiffs filed their First Amended Complaint. The First Amended Complaint added several members of the Board of Regents as Defendants, alleging that they were ultimately responsible for the alleged acts of infringement at GSU because of their supervisory authority over the University System of Georgia. Defendants’ Answer to the First Amended Complaint again denied infringement, asserted fair use, and claimed sovereign immunity and Eleventh Amendment immunity for all Defendants.
In late December 2008, the University System of Georgia convened a Select Committee on Copyright to review GSU’s then-existing copyright policy, which was called the “Regents’ Guide to Understanding Copyright & Educational Fair Use.”8 On February 17, 2009, the Select Committee announced a new copyright policy for GSU (the “2009 Policy”), which went into effect the same day. Under the 2009 Policy, a revised version of which remains in effect today, GSU professors who wish to post an excerpt of a copyrighted work on ERes or uLearn for distribution to their students must first determine whether they believe that doing so would be fair use. In order to make this determination, professors must fill out a “Fair Use Checklist” for each excerpt.9
The Checklist allows GSU professors to perform a version of the analysis a court *1243might perform should the professor claim fair use in a subsequent copyright infringement suit. As described above, see supra part I.A, the fair use analysis involves a consideration of whether allowing the unpaid use in a given case would be equitable and serve the objectives of copyright in light of four statutory factors, see 17 U.S.C. § 107. For each factor, the Checklist provides several criteria that purportedly weigh either for or against a finding of fair use, each with a corresponding checkbox.10 The Checklist instructs professors to check each criterion that applies, and then add up the checks to determine whether the factor weighs in favor of or against a finding of fair use. After making this tally, the Checklist explains that “[wjhere the factors favoring fair use outnumber those against it, reliance on fair use is justified. Where fewer than half the factors favor fair use, instructors should seek permission from the rights holder.” Doc. 235-2, at 7. Thus, under the 2009 Policy, a GSU professor may post an excerpt of a copyrighted work on ERes or uLearn without obtaining a permission from the copyright holder if the professor first decides that doing so would be protected by the doctrine of fair use, according to the criteria set forth in the Checklist.
After completing an initial round of discovery, both parties moved for summary judgment on February 26, 2010. Plaintiffs alleged that the 2009 Policy had failed to curb the alleged infringement of their copyrighted works, and argued that they were entitled to summary judgment on all claims. Plaintiffs argued that they were entitled to an injunction based on the alleged infringements listed in their First Amended Complaint, that had occurred prior to enactment of the 2009 Policy, and added allegations regarding new instances of infringement, which also occurred prior to enactment of the 2009 Policy. Plaintiffs also argued that injunctive relief was appropriate as to Defendants under Ex parte Young, 209 U.S. 123, 28 S.Ct. 441, 52 L.Ed. 714 (1908), which permits prospective injunctive relief against state officers in their official capacities.
Defendants argued that they were entitled to summary judgment on all claims. Defendants also contended that Plaintiffs were only entitled to prospective declaratory and injunctive relief as to ongoing and continuous conduct, and so the District Court should only consider alleged infringements that occurred since GSU enacted the 2009 Policy. Thus, Defendants contended, any claims based on GSU’s superseded pre-2009 policy were moot.. Defendants also claimed that GSU’s adoption of the 2009 Policy had substantially reduced the amount of Plaintiffs’ works used by GSU professors.
On August 11, 2010, and August 12, 2010, the District Court ordered Plaintiffs to produce a list of all claimed infringements of their works that had occurred at GSU during the period following enactment of the 2009 Policy, which included the 2009 “Maymester” (a three-week *1244term), the 2009 summer semester, and the 2009 fall semester. The District Court also required that Plaintiffs provide certain information about the alleged infringements, including the name of the course in which the excerpt of the work was used, the instructor’s name, a title and description of the copied work, the name of the owner of the work’s copyright, the number of pages and chapters in the work, the number of pages and chapters copied, the retail price to purchase the entire work, and the potential cost to license each excerpt at issue. On August 20, 2010, Plaintiffs filed a list showing 126 claimed infringements.
On September 30, 2010, the District Court denied Plaintiffs’ motion for summary judgment. The District Court granted in part and denied in part Defendants’ motion for summary judgment, granting summary judgment to Defendants on the claims of direct and vicarious infringement, and denying summary judgment to Defendants on the claim of contributory infringement.
In its summary judgment order, the District Court construed the First Amended Complaint as claiming that the 2009 Policy as applied to Plaintiffs’ works “has led to continuing abuse of the fair use privilege.” Doc. 235, at 5. Accordingly, the District Court agreed that only alleged acts of infringement that took place after GSU enacted the 2009 Policy were relevant to Plaintiffs’ claims, and held that it would only consider those acts. However, the District Court declined to consider at that time the list of alleged post-2009 Policy infringements that Plaintiffs had provided because the parties had not yet conducted discovery as to these allegations.
The District Court granted summary judgment for Defendants on the direct infringement claim because it found that GSU as an entity is not itself capable of copying or making fair use determinations, and that Defendants cannot be held liable under a respondeat superior theory because “respondeat superior applies in the copyright context as a basis for finding vicarious liability, not direct liability.” Id. at 19. With regard to the vicarious infringement claim, the District Court held that summary judgment for Defendants was appropriate because it found that there was no evidence to support the conclusion that GSU had profited from the allegedly infringing use of Plaintiffs’ works by GSU personnel, or that GSU’s shift to digital distribution of excerpts via its electronic reserve system served as a “direct draw for students or that Defendants have a direct financial interest in copyright infringement.” Id. at 21. With regard to the contributory infringement claim, the District Court denied both Plaintiffs’ and Defendants’ motions for summary judgment, finding that the record was silent as to whether the 2009 Policy “encourage[s] improper application of the fair use defense.” Id. at 30.
The District Court further held:
Going forward, in order to show that Defendants are responsible for the copyright infringements alleged in this case, Plaintiffs must show that the 2009 Copyright Policy resulted in ongoing and continuous misuse of the fair use defense. To do so, Plaintiffs must put forth evidence of a sufficient number of instances of infringement of Plaintiffs’ copyrights to show such ongoing and continuous misuse. Defendants will have the burden of showing that each specified instance of 2009 Copyright Policy infringement was a fair use.
Both sides will be limited to the list of claimed infringements produced in response to the Court’s August 11, 2010 and August 12, 2010 orders.
Id. at 30.
On October 20, 2010, Plaintiffs filed a motion for partial reconsideration of the *1245District Court’s grant of summary judgment for Defendants. Plaintiffs argued, in relevant part, that the District Court erred in dismissing the direct infringement claim because courts routinely hold employers responsible for acts of copyright infringement committed by their employees through the doctrine of respondeat superior without recourse to secondary infringement doctrines. The District Court granted Plaintiffs’ motion, reviving Plaintiffs’ claim of direct infringement, striking Plaintiffs’ label of the claim as “direct infringement,” and construing the claim as an “indirect infringement” claim.11
On November 4, 2010, Defendants moved to dismiss Plaintiffs’ First Amended Complaint for lack of subject matter jurisdiction. Defendants argued that Plaintiffs’ claims were barred by the Eleventh Amendment, and that Plaintiffs could not make use of the Ex parte Young exception to Eleventh Amendment immunity because Defendants were not personally violating — or threatening to violate — copyright law. Defendants contended that their supervisory role over GSU’s policies and personnel did not create a sufficient causal nexus to the infringing activity, and thus Ex parte Young did not apply. The District Court denied Defendants’ motion to dismiss, holding that it would not presently address Defendants’ argument that Ex parte Young did not apply because “Plaintiffs’ First Amended Complaint currently alleges that Defendants’ own copying, scanning, displaying, and distributing of Plaintiffs’ materials violated Plaintiffs’ copyrights.” Doc. 267, at 12.
On November 5, 2010, the District Court ordered discovery regarding GSU’s use of Plaintiffs’ copyrighted material during the 2009 Maymester, the 2009 summer term, and the 2009 fall term. In response, on March 15, 2011, the parties filed a'joint document detailing ninety-nine alleged acts of infringement that occurred during that period. This filing included the name of each allegedly infringed work, the circumstances of the alleged infringement, and Defendants’ objections to each claimed infringement. For all ninety-nine, Defendants claimed fair use. In some instances, Defendants also argued that Plaintiffs had not provided evidence of a copyright registration for the work in question. Defendants also objected to the method by which Plaintiffs had calculated the ratio between the number of pages in an excerpt and the total number of pages in the work. Plaintiffs had made these calculations based on the total number of pages in the textual body of the work, excluding parts such as tables of contents, indexes, and prefaces. Defendants argued that these non-body parts of a work should be included in the total page count of a work.
On May 11, 2011, in response to the District Court’s Consolidated Pretrial Order dated May 2, 2011, Plaintiffs filed a proposed order for injunctive relief. Plaintiffs’ proposed injunction would essentially limit GSU to copying that fell within the parameters set forth in the Agreement on Guidelines for Classroom Copying in Not-For-Profit Educational Institutions with Respect to Books and Periodicals (the “Classroom Guidelines”).12 See Doc. 300-1, at 3-4.
*1246A bench trial began on May 17, 2011. At the close of Plaintiffs’ case, the District Court granted Defendants’ motion for judgment on the contributory infringement claim, leaving only the claim that the 2009 Policy directly caused copyright infringement.
*1247On June 1, 2011, Plaintiffs voluntarily filed a revised list of seventy-five claimed infringements, relating to sixty-four separate works. This list dropped twenty-five of the claimed infringements from the joint list filed on March 15, 2011, and added one new claim. The District Court accepted Plaintiffs’ revised list, including the added claim, because it found that the added claim featured the same copied material as a previously identified claim, and that both sides had an adequate opportunity to address it at trial. Cambridge Univ. Press, 863 F.Supp.2d at 1204-05 n. 7. However, the District Court construed two claims of infringement involving the same copied material in two different semesters as one claim, because the evidence showed that the material was posted only once for the same multi-semester course, and that the material was accessed during only one semester. Id. at 1289 n. 89. Thus, in reaching its decision, the District Court considered a total of seventy-four individual claims of infringement.
On May 11, 2012, the District Court issued an order holding that Defendants had infringed Plaintiffs’ copyright in five of the seventy-four instances at issue. Id. at 1363. First, the District Court held that Plaintiffs were entitled to the benefit of the Ex parte Young exception to Eleventh Amendment immunity because the GSU Defendants were responsible for the creation and implementation of the 2009 policy, and the Board of Regents at least tacitly approved it. Id. at 1209.
After holding that it could appropriately enter injunctive relief pursuant to Ex parte Young, the District Court described how it would proceed. First, because “there is no precedent on all fours for how the factors should be applied where excerpts of copyrighted works are copied by a nonprofit college or university for a nonprofit educational purpose,” the District Court would determine how the fair use factors should be applied in this case, and whether any other factors merit consideration. Id. at 1210. Then, in order to assess the efficacy of GSU’s copyright policy in complying with the requirements of the law, the District Court would assess each of the seventy-four claimed instances of infringement individually and “compare the outqome of this process to the outcomes that were achieved under the [fair use] checklists prescribed under the 2009 Copyright Policy.” Id. at 1211.
First, the District Court made several findings of fact. Id. at 1211-21. In addition to’ findings — described above — regarding GSU’s copyright policy, Plaintiffs’ operations, and CCC’s services for licensing of excerpts of copyrighted works, the District Court made detailed findings regarding the market for excerpts of Plaintiffs’ works, based largely on the parties’ stipulations.
With regard to CCC’s APS and ECCS programs, the District Court noted that publishers determine the fee for licensing excerpts of their materials, which for academic users is currently between ten and twenty-five cents per page. Id. at 1215. Cambridge stipulated that it charges eleven cents per page for APS and fifteen cents per page for ECCS, Sage stipulated that it charges fourteen cents per page (unspecified whether for permissions via APS, ECCS, or both), and Oxford stipulated that it charges twelve cents per page (unspecified whether for APS, ECCS, or both). Id. CCC charges a $3.50 service charge per order, and keeps 15 percent of the permission fee. Id. All three Plaintiffs have in-house permissions departments, but 90 percent of Oxford’s permissions income and 95 percent of Cambridge’s comes from CCC. Id. Although it did not note a percentage, the District Court found that “many though not all of Sage’s works at *1248issue in this case generated permissions revenue from CCC during the period from July 1, 2004 to December 1, 2010.” Id. at 1214.
The District Court found that “Plaintiffs earn considerable annual rights and permissions income through CCC. CCC made rights and permissions payments to Cambridge, Oxford, and Sage totaling $4,722,686.24 in FY 2009 and $5,165,445.10 in FY 2010,” amounting to an average of $1,574,228.74 per-Plaintiff in FY 2009. Id. at 1216 (footnote omitted). However, these totals involve payments through services which have “no demonstrated relevance to this case.” Id.
Considering only payments for permissions under the academic permissions services — the APS, ECCS, and Academic Annual Copyright License13 programs — in FY 2009 Cambridge earned $404,494.90, Oxford earned $480,622.47, and Sage earned $352,759.84. Id. APS income, which “is dominated by payments for printed coursepacks,” accounted for the majority of these totals. Id. Thus, in FY 2009, through ECCS — CCC’s academic permissions service for individual licensing of digital copies of excerpts — Cambridge earned $81,671.35, Oxford earned $70,485.81, and Sage earned $85,660.91. Id. The figures for FY 2010 are similar. Id.
The District Court calculated the percentage of Plaintiffs’ total revenues that these figures represent and determined that
on average, APS and ECCS permissions represent .0024 (one-quarter of one percent) of net revenue in FY 2009 for each of the Plaintiffs. If the calculation is limited to ECCS income [for licensing of digital excerpts] (an average of $79,272 per Plaintiff), the percentage would be .00046 (five one-hundredths of one percent) of average net revenues. Even if all of the types of permissions payments reflected in Plaintiffs’ referenced exhibits are included, this income would represent an average of .0093 (nine-tenths of one percent) of net revenues per Plaintiff for FY 2009.
Id.
The District Court noted that Plaintiffs offered no testimony or evidence demonstrating that they lost book sales on account of Defendants’ actions, and accordingly found that no book sales were lost. Id. at 1217. The District Court also found that if GSU students had been required to pay permissions fees for the use of excerpts of Plaintiffs’ works in 2009, “there would have been some small overall increase in the cost of education.” Id.
The District Court turned to its conclusions of law, first laying out the requirements for establishing a prima facie case of copyright infringement. To establish a prima facie case, Plaintiffs must establish that (1) they own valid copyrights in the allegedly infringed works and (2) that Defendants copied protected elements of the allegedly infringed works. Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340, 361, 111 S.Ct. 1282, 1296, 113 L.Ed.2d 358 (1991). Plaintiffs must also establish that the copyright for the allegedly infringed work has been registered in the U.S. Copyright Office. 17 U.S.C. § 411(a).
The District Court found that there were valid copyright registrations in all but one of the instances of alleged infringement. Cambridge Univ. Press, 863 *1249F.Supp.2d at 1221. The District Court also found that Plaintiffs had established the element of copying protected material in all instances of alleged infringement. Id. With regard to the ownership prong, the District Court found that Plaintiffs are generally assignees of copyrights originally owned by authors, or hold exclusive licenses to publish the works in question pursuant to contracts with authors. Id. at 1222. Sage is deemed to be the author of some of its edited books by virtue of work-for-hire contracts. Id. However, the District Court found that several instances of alleged infringement involved works for which the Plaintiff-publisher offered no evidence of a contract with the contributing author. Id. The District Court held that, because such evidence is necessary in this case to establish ownership of the copyright to the works in question, in those instances lack of such evidence would be fatal to Plaintiffs’ prima facie case of infringement. Id. at 1223.
Then, the District Court set forth the parameters of its fair use analysis. The District Court held that the first fair use factor, “the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes,” 17 U.S.C. § 107(1), “strongly favor[ed] Defendants” in all instances because “[t]his ease involves making copies of excerpts of copyrighted works for teaching students and for scholarship ... [and so] [t]he use is for strictly nonprofit educational purposes,” Cambridge Univ. Press, 863 F.Supp.2d at 1224-25.
The District Court held that the second fair use factor, “the nature of the copyrighted work,” 17 U.S.C. § 107(2), favored Defendants in all instances because it found — after undertaking an individualized review of all of the works at issue for which it found that Plaintiffs had made a prima facie case of infringement — that “the books involved in this case are properly classified as informational in nature, within the spectrum of factual materials and hence favoring fair use.” Cambridge Univ. Press, 863 F.Supp.2d at 1226.
The District Court held that the third fair use factor, “the amount and substan-tiality of the portion used in relation to the copyrighted work as a whole,” 17 U.S.C. § 107(3), “favor[ed] either Plaintiffs or Defendants, depending on the amount taken from each book,” Cambridge Univ. Press, 863 F.Supp.2d at 1235.
After hearing testimony from several GSU professors as to the reasoning behind their choice of a particular excerpt and how use of that excerpt furthered the professor’s goals for a particular class, the District Court found that all of the selections furthered the legitimate educational purposes of the courses in which they were used. Id. The District Court also found that some professors’ educational purposes were furthered by using whole chapters of books, because chapters typically contain a complete treatment of a topic. Id. at 1234
The District Court then determined that “[t]he right approach is to select a percentage of pages which reasonably limits copying and to couple that with a reasonable limit on the number of chapters which may be copied.” Id. at 1235. Accordingly, the District Court held that
[w]here a book is not divided into chapters or contains fewer than ten chapters, unpaid copying of no more than 10 percent of the pages in the book is permissible under factor three.... Where a book contains ten or more chapters, the unpaid copying of up to but no more than one chapter (or its equivalent) will be permissible under fair use factor three.... The chapter or other excerpt must fill a demonstrated, legitimate purpose in the course curriculum and must *1250be narrowly tailored to accomplish that purpose. Where the foregoing limitations are met factor three will favor fair use, i.e., will favor Defendants. Otherwise factor three will favor Plaintiffs.
Id. at 1243.
The District Court held that its factor three analysis would not be bound by the standards provided in the Classroom Guidelines. Id. at 1227-29. The District Court also chose to consider non-body material — such as dedications and introductions — as part of the work for purposes of calculating the work’s total page count, noting that such material constitutes copyrightable expression, and rejected Plaintiffs’ argument — raised late in the proceedings, that Defendants’ took 100 percent of a work when copying a whole chapter from a book of chapters by individual authors on individual subjects — on grounds of untimeliness and unfair surprise. Id. at 1230-31.
With regard to the fourth fair use factor, “the effect of the use upon the potential market for or value of the copyrighted work,” 17 U.S.C. § 107(4), the District Court found that “Defendants’ use of small excerpts did not affect Plaintiffs’ actual or potential sales of books” because they do not substitute for the books, Cambridge Univ. Press, 863 F.Supp.2d at 1236. However, the District Court found that Defendants’ use of excerpts may be at the cost of Plaintiffs’ licensing revenues, and so may affect the market for licensing of excerpts. Id. at 1237. Thus, the District Court concluded that, in reviewing the individual instances of alleged infringement, it would analyze fair use factor four as follows:
[Wjhere permissions are readily available from CCC or the publisher for a copy of a small excerpt of a copyrighted book, at a reasonable price, and in a convenient format (in this case, permissions for digital excerpts), and permissions are not paid, factor four weighs heavily in Plaintiffs’ favor. Factor four weighs in Defendants’ favor when such permissions are not readily available.
Id. at 1243.
However, “[bjecause Plaintiffs advocate that CCC has created an effective means through which Defendants could have obtained licensed copies of the excerpts in question here, the [District] Court place[d] the burden on Plaintiffs to show that CCC provided, in 2009, reasonably efficient access to the particular excerpts involved in this case.” Id. at 1237. Thus, in cases where no evidence showed whether digital permissions were readily available for excerpts of a particular work, the District Court found that the fourth fair use factor favored Defendants.
The District Court took into account two additional considerations. First, the District Court noted that, based on testimony that “royalties are not an important incentive for academic writers,” and on a presumption that that academic authors publish primarily to enhance their professional reputation and contribute to academic knowledge, “[tjhere is no reason to believe that allowing unpaid, nonprofit academic use of small excerpts in controlled circumstances would diminish creation of academic works.” Id. at 1240. Second, the District Court found that “it is consistent with the principles of copyright to apply the fair use doctrine in a way that promotes the dissemination of knowledge, and not simply its creation.” Id. at 1241. The District Court noted that the evidence demonstrates that academic permissions income does not represent a significant portion of Plaintiffs’ overall revenue. Id. Thus, the District Court found that a slight diminution of Plaintiffs’ permissions income caused by the District Court’s findings of fair use would not appreciably harm Plaintiffs’ ability to publish scholarly *1251works. Id. at 1243. On the other hand, the District Court found, “[mjaking small free excerpts available to students would further the spread of knowledge.” Id. at 1242.
The District Court proceeded to undertake an individualized analysis of the seventy-four individual instances of alleged infringement, subject to the foregoing analysis. Id. at 1243-1363. The District Court held that Plaintiffs had established a prima facie case of copyright infringement in forty-eight instances. Id. at 1254-59; 1263; 1265-67; 1270-77; 1281; 1283-1296; 1299-1300; 1303-1310; 1316; 1322-24; 1329-30; 1333; 1339-51; 1355-62. For the other twenty-six instances, the District Court held that Plaintiffs had failed to establish a prima facie case either because Plaintiffs failed to demonstrate ownership of or (in one case) a valid registration of a copyright in the work in question, or because any infringement was de minimis in cases where the evidence showed that students either never accessed the excerpt or were required to buy the full book from which the excerpt in question had been copied. Id. at 1245-53; 1262; 1265; 1269; 1280; 1283; 1299; 1302; 1312-15; 1319-21; 1326-27; 1332; 1335-37; 1354.
In weighing the fair use factors to assess each of the forty-eight instances of alleged infringement for which the District Court found that Plaintiffs had established a prima facie case, the District Court held that fair use applied whenever at least three of the four factors favored Defendants. Id. at 1255-58; 1264; 1266; 1271-76; 1282; 1284-97; 1300-01; 1304-11; 1317; 1323-25; 1329-31; 1334; 1339-52; 1356; 1361. Because the District Court found that factors one and two favored Defendants in all cases, the District Court essentially held that fair use applied each time a professor posted an excerpt that fell within the 10 percent-or-one-chapter limit on allowable copying the District Court had set (such that factor three favored Defendants) and each time there was no evidence that digital permissions were available for excerpts of the work in question (such that factor four favored Defendants).
With regard to factor three, in thirty-five of the forty-eight claims of infringement, the District Court found that the copying was “decidedly small” because it fell within the 10 percent-or-one-chapter limit, and so factor three favored Defendants. Id. at 1254-58; 1266; 1270-72; 1276; 1281; 1283-88; 1291; 1295-97; 1300-01; 1304-08; 1317; 1325; 1329-30; 1334; 1339-45; 1348-50; 1356; 1361. In the other thirteen cases, the copying exceeded the 10 percent-or-one-chapter limit, and so the District Court held that factor three favored Plaintiffs. Id. at 1259; 1263; 1268; 1274; 1277; 1290; 1294; 1310; 1323; 1346; 1352; 1358; 1362.
With regard to factor four, in seventeen of the forty-eight cases, the District Court found that the parties had presented no evidence regarding licensing availability, but because the District Court placed the burden on this issue on Plaintiffs, the District Court found that factor four favored Defendants. Id. at 1290-97; 1300-01; 1311; 1317; 1323-25; 1329-31; 1340-42; 1347; 1352. In the other thirty-one cases, the District Court found that Plaintiffs had made digital licensing available for excerpts of the work in question, and so, because there was a “ready market for licensed digital excerpts of [the] work in 2009,” factor four strongly favored Plaintiffs. Id. at 1255-60; 1263; 1266-68; 1271-78; 1281; 1284-88; 1304-09; 1334; 1339; 1344-45; 1348-50; 1356-62.
When the District Court’s weighing of the factors resulted in a tie — factors one and two favored Defendants and factors three and four favored Plaintiffs — the Dis*1252trict Court revisited its analysis of the factors, reviewing and reweighing the importance of the factors for the specific work in question. Id. at 1260; 1263-64; 1268-69; 1274; 1278; 1358-59; 1363. The District Court performed this reweighing for seven claims of infringement. Id. In two instances, the District Court held that fair use applied because Plaintiffs’ revenues from digital permissions were sufficiently small as a proportion of total revenues to render factor four of less importance. Id. at 1263-64; 1274. The District Court held that fair use did not apply in the remaining five instances. Id. at 1260; 1268-69; 1278; 1358-59; 1363.
Thus, of the forty-eight instances of alleged infringement for which the District Court found that Plaintiffs had established a prima facie case, the District Court held that Defendants had infringed Plaintiffs’ copyrights in five instances, id. at 1260; 1269; 1278; 1358-59; 1363, and that the fair use defense applied in forty-three, id. at 1255-58; 1264; 1266; 1271-76; 1282; 1284-97; 1300-01; 1304-11; 1317; 1323-25; 1329-31; 1334; 1339-52; 1356; 1361. The District Court concluded that the 2009 Policy had caused the five instances of infringement. Id. at 1363. In reaching this conclusion, the District Court noted that the 2009 Policy did not limit copying to excerpts which were “decidedly small,” did not prohibit the copying of multiple chapters from the same book, and did not provide sufficient guidance in determining the effect the use of an excerpt may have on the market for or value of the copyrighted work. Id.
In light of its findings, the District Court directed Plaintiffs to propose appropriate injunctive and declaratory relief. Id. at 1364. On May 31, 2012, Plaintiffs filed a second proposed injunction. This injunction would allow copying without permission in a manner that appears to be modelled on the fair use parameters set forth in District Court’s judgment, as described above. The proposed injunction would also require a reasonable investigation as to license availability, recordkeeping as to unlicensed copying and the related license availability investigations, periodic reports and certifications of compliance by Defendants to the District Court for a three-year period, and periodic granting of access to GSU’s electronic reserve systems to Plaintiffs’ representative over a three-year period.
In their Opposition to Plaintiffs’ Request for Injunctive Relief, filed on June 15, 2012, Defendants stated that GSU had revised its 2009 Policy in accordance with the District Court’s May 11, 2012, order, and submitted into evidence revised versions of the 2009 Policy and Fair Use Checklist which purportedly “incorporate[d] the provisions of the Court’s Order that were not already present in the 2009 Copyright Policy.” Doc. 432, at 15; Doc. 432-1; Doc. 432-2.
On August 10, 2012, the District Court issued an order entering declaratory and injunctive relief. The District Court enjoined Defendants “to maintain copyright policies for Georgia State University which are not inconsistent with the Court’s Order of May 11, 2012 and this Order.” Doc. 441, at 11. By way of declaratory relief, the District Court clarified the holding of its May 11, 2012, order in three detailed sections, titled as follows:
A. The requirement that excerpts be “decidedly small” to tip factor three in Defendants’ favor applies to the aggregate of all excerpts from a book which are assigned during the term of the course.
B. The holdings of this case do not address fair use of books intended *1253solely for instruction of students enrolled in a class.
C. Fair use protection is conditioned on strict reliance with measures calculated to protect copyrighted excerpts from unwarranted distribution.
Id. at 2-9.
Finally, because “Defendants prevailed on all but five of the [ninety-nine] copyright claims which were at issue when the trial began,” the District Court declared Defendants the prevailing party, and determined that it would exercise its discretion to award attorneys’ fees and costs to Defendants. Id. at 12-13. The District Court noted:
When the trial began, Plaintiffs chose to pursue 99 claims out of 126. They then dropped 25 claims (and added one) during the trial. As to the remaining 75 claims, no prima facie case was proven in 26 instances. Digital permissions were unavailable in 33 instances. Neither digital nor hard copy permissions were available in 18 eases.... Plaintiffs’ failure to narrow their individual infringement claims significantly increased the cost of defending the suit.
Id. at 14. On September 10, 2012, Plaintiffs appealed the District Court’s August 10 order entering declaratory and injunctive relief, seeking review of that order and all prior orders entered.
Finally, on September 30, 2012, the District Court issued an order awarding to Defendants attorneys’ fees in the amount of $2,861,348.71 and costs in the amount of $85,746.39; and entered final judgment on its finding of infringement, its order granting declaratory and injunctive relief, and the award of fees and costs to Defendants. On October 2, 2012, Plaintiffs appealed the District Court’s September 30 order and final judgment, as well as all prior orders. On October 23, 2012, the District Court stayed execution of the portion of its judgment awarding Defendants attorneys’ fees and costs pending appeal.
We consolidated Plaintiffs’ appeals. On appeal, Plaintiffs do not contest the District Court’s ruling that Plaintiffs failed to make out a prima facie case of infringement in twenty-six instances of copying. However, Plaintiffs argue that the District Court’s application of the fair use factors was legally flawed, and that the District Court consequently erred in finding that the fair use defense applied in forty-three of the forty-eight remaining instances of alleged infringement.14
Furthermore, Plaintiffs argue, because the District Court erred in identifying only five instances of infringement during the period it examined in 2009 and failed to apprehend that the 2009 Policy facilitates widespread, ongoing infringement, the injunction the District Court ordered is “unduly narrow.” ' Appellants’ Br. at 83. Thus, Plaintiffs contend, we must vacate the injunction “and remand with instructions to enter an appropriately comprehensive injunction]” along the lines of the injunctive relief Plaintiffs proposed below.15 Id.
*1254Plaintiffs also argue that the District Court erred in awarding attorney’s fees and costs to Defendants. Plaintiffs argue the District Court should have considered the claims of infringement as a whole instead of conducting a work-by-work analysis. Plaintiffs contend that the District Court’s “erroneous treatment of the individual claims [of infringement] ... produced a misleading tally of unsuccessful and successful claims which obscured the fact that the court held GSU’s [2009] [Policy unlawful for allowing excessive copying.” Id. at 84. The relevant measure of success is not, Plaintiffs contend, which party prevailed on the most individual claims, but rather whether or not the 2009 Policy caused copyright infringement. Because the District Court held that the 2009 Policy did cause copyright infringement, Plaintiffs claim that the District Court erred in designating Defendants as the prevailing party for the purpose of awarding fees and costs. Even if the District Court did not err in designating Defendants as the prevailing party, Plaintiffs further argue, the District Court abused its discretion in awarding fees and costs to Defendants because it made the award despite finding that Plaintiffs acted in “good faith in bringing ... suit” and that there was “no controlling authority governing fair use in a nonprofit educational setting.” See Doc. 441, at 14. Thus, Plaintiffs contend, we must reverse the District Court’s ruling as to Defendants’ entitlement to attorneys’ fees and costs. Finally, Plaintiffs argue that even if we allow the award of fees and costs to stand, we must reduce the award by $142,038.54, an amount awarded for fees paid to Defendants’ expert, Dr. Kenneth Crews, because, Plaintiffs’ argue, expert witness fees are not recoverable.
Defendants argue that the District Court did not err in its fair use analysis. Thus, Defendants contend, in light of their successful defense in all but five of the alleged instances of infringement, the District Court’s injunction was proper. Defendants also argue that Plaintiffs’ more restrictive proposed injunction is unnecessary because GSU already revised its copyright policy in order to comply with the District Court’s order. Accordingly, Defendants contend, we must affirm the District Court’s order finding five instances of infringement and its entry of injunc-tive and declaratory relief.
Defendants argue that the District Court properly undertook a work-by-work analysis of the individual claims of infringement and did not err in concluding that Defendants were the prevailing party because disposition of the individual instances of infringement represents the central issue in the case, upon which Plaintiffs did not prevail, having proved only five instances of infringement out of the ninety-nine Plaintiffs alleged at the beginning of trial. Because Plaintiffs refused to limit the scope of their claims, which significantly increased Defendants’ costs, Defendants argue that the District Court did not abuse its discretion in awarding fees and costs to Defendants. Defendants also argue that the District Court did not err in awarding expert witness fees to Defendants, because such fees are routinely awarded as incidental expenses incurred as part of attorneys’ work on litigation. Accordingly, Defendants contend, we must affirm the District Court’s award of attorneys’ fees and costs.
Finally, Defendants argue — apparently in the alternative to Defendants’ argument that the District Court’s entry of injunctive and declaratory relief should be affirmed— that, because Plaintiffs cannot establish that the 2009 Policy violates any federal law, and because Defendants’ generalized responsibility for GSU policy is not sufficient to establish the necessary connection *1255to GSU professors’ infringing actions, Defendants are immune from suit pursuant to the Eleventh Amendment and the Ex parte Young exception does not apply. Plaintiffs counter that because Defendants did not raise this argument on cross-appeal we should not consider it.
II.
“After a bench trial, we review [a] district court’s conclusions of law de novo and [a] district court’s factual findings for clear error.” Proudfoot Consulting Co. v. Gordon, 576 F.3d 1223, 1230 (11th Cir.2009). Fair use involves both questions of law and questions of fact. Harper Row Publishers, Inc. v. Nation Enters., 471 U.S. 539, 560, 105 S.Ct. 2218, 2230, 85 L.Ed.2d 588 (1985).
We review a district court’s decision to award injunctive relief for abuse of discretion. Klay v. United Healthgroup, Inc., 376 F.3d 1092, 1096 (11th Cir.2004); see also Pac. & S. Co. v. Duncan, 744 F.2d 1490, 1499 (11th Cir.1984) (reviewing for abuse of discretion the district court’s decision not to permanently enjoin a seller of videotapes containing material that the court had held infringed copyrights).
We review de novo a district court’s determination as to whether a party is the “prevailing party” and hence eligible for an award of fees under the Copyright Act, 17 U.S.C. § 505. Dionne v. Floormasters Enters., Inc., 667 F.3d 1199, 1203 (11th Cir.2012). We review a district court’s ultimate decision to award attorneys’ fees or costs for abuse of discretion. Id.
III.
A.
As an initial matter, we must dispose of Defendants’ contention that they are immune from suit pursuant to the Eleventh Amendment. Because this argument is outside the scope of Plaintiffs’ brief and was not raised by Defendants on cross-appeal, we find that the argument is not properly raised. As we have previously explained, “a party who has not appealed may not bring an argument in opposition to a judgment or attack the judgment in any respect, or hitch a ride on his adversary’s notice of appeal to enlarge his rights under the judgment or diminish those of the opposing party.” Lawhorn v. Allen, 519 F.3d 1272, 1286 n. 20 (11th Cir.2008) (citations omitted) (quotation marks omitted). An argument that the District Court’s ruling must be vacated because suit is barred by Eleventh Amendment immunity is not excepted from this rule. Majette v. O’Connor, 811 F.2d 1416, 1419 n. 3 (11th Cir.1987). Accordingly, we decline to address Defendants’ Eleventh Amendment argument.16
B.
The Copyright Clause of the U.S. Constitution provides that Congress shall have the power “[t]o promote the Progress of Science17 ... by securing for limited *1256Times to Authors ... the exclusive Right to their respective Writings.... ” U.S. Const., Art. I, § 8, cl. 8. As the Supreme Court has explained, “the economic philosophy behind the clause empowering Congress to grant copyrights is the conviction that encouragement of individual effort by personal gain is the best way to advance public welfare [by promoting the creation and dissemination of ideas] through the talents of authors.” Mazer v. Stein, 347 U.S. 201, 219, 74 S.Ct. 460, 471, 98 L.Ed. 630 (1954).
Promoting the creation and dissemination of ideas has been the goal driving Anglo-American copyright law since the enactment of the first English copyright statute to explicitly vest copyright in a work’s creator, the Statute of Anne of 1710, which declared that it was “[a]n Act for the Encouragement of Learning, by Vesting the Copies of Printed Books in the Authors ... during the Times therein mentioned.” 8 Ann., c. 19 (1710); see also Pierre N. Leval, Toward a Fair Use Standard, 103 Harv. L.Rev. 1105 (1990) (describing the Statute of Anne and its influence on early U.S. copyright law and the fair use doctrine). Thus, in our tradition, “copyright is not an inevitable, divine, or natural right that confers on authors the absolute ownership of their creations. It is designed rather to stimulate activity and progress in the arts for the intellectual enrichment of the public.” Cariou v. Prince, 714 F.3d 694, 705 (2d Cir.2013) (quoting Leval, supra, at 1107); see also Aiken, 422 U.S. at 156, 95 S.Ct. at 2044 (“The immediate effect of our copyright law is to secure a fair return for an ‘author’s’ creative labor. But the ultimate aim is, by this incentive, to stimulate artistic creativity for the general public good.”).
The Copyright Act furthers this purpose by granting authors a bundle of “exclusive rights,” 17 U.S.C. § 106, “in original works of authorship fixed in any tangible medium of expression,” id. § 102, for a limited time, id. §§ 302-305. While an author holds a copyright in his or her work, the author may control, for example, reproduction of the work or distribution of the work to the public. Id. § 106(1), (3).
In part because copyright is not grounded in authors’ natural rights but rather meant to provide maximal public benefit, the Copyright Act’s grant to authors of a monopoly over the use of their works is limited in several important ways beyond its finite duration.18 Golan v. *1257Holder, — U.S. - — , 132 S.Ct. 873, 890-91, 181 L.Ed.2d 835 (2012). For example, “[c]opyright cannot protect an idea, only the expression of that idea.” Suntrust Bank v. Houghton Mifflin Co., 268 F.3d 1257, 1263 (11th Cir.2001); see 17 U.S.C. § 102(b) (“In no case does copyright protection for an original work of authorship extend to any idea, procedure, process, system, method of operation, concept, principle, or discovery, regardless of the form in which it is described, explained, illustrated, or embodied in such work”). Thus, “copyright assures authors the right to their original expression, but encourages others to build freely upon the ideas and information conveyed by a work.” Feist, 499 U.S. at 349-50, 111 S.Ct. at 1290.
The fair use doctrine also critically limits the scope of the monopoly granted to authors under the Copyright Act in order to promote the public benefit copyright is intended to achieve. See Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 575, 114 S.Ct. 1164, 1169, 127 L.Ed.2d 500 (1994) (“From the infancy of copyright protection, some opportunity for fair use of copyrighted materials has been thought necessary to fulfill copyright’s very purpose, ‘[t]o promote the Progress of Science-’” (quoting U.S. Const., Art. I, § 8, cl. 8.)). In other words, fair use provides necessary “breathing space within the confines of copyright.” Id. at 579, 114 S.Ct. at 1171. By allowing for the limited use of copyrighted works without the permission of the copyright holder by members of the public in certain circumstances, fair use “permits [and requires] courts to avoid rigid application of the copyright statute when, on occasion, it would stifle the very creativity which that law is designed to foster.” Id. at 577, 114 S.Ct. at 1170 (alteration in original) (quoting Stewart v. Abend, 495 U.S. 207, 236, 110 S.Ct. 1750, 1767, 109 L.Ed.2d 184 (1990)).
In a sense, the grant to an author of copyright in a work is predicated upon a reciprocal grant to the public by the work’s author of an implied license for fair use of the work. See Harper & Row, 471 U.S. at 549, 105 S.Ct. at 2225 (“[T]he author’s consent to a reasonable use of his copyrighted works ha[d] always been implied by the courts as a necessary incident of the constitutional policy of promoting the progress of science ... since a prohibition of such use would inhibit subsequent writers from attempting to improve upon prior works and thus ... frustrate the very ends sought to be attained.” (quoting H. Ball, Law of Copyright and Literary Property 260 (1944))). Thus, in order to promote the creation of new works, our laws contemplate that some secondary users — those implied licensees making fair use of copyrighted works — will be allowed to make use of original authors’ works. At the same time, a secondary user who takes overmuch in the name of fair use operates outside the bounds of his of her implied-by-law license.
How much unpaid use should be allowed is the bailiwick of the fair use doctrine. To further the purpose of copyright, we must provide for some fair use taking of copyrighted material. This may be viewed as a transaction cost, incidental to the business of authorship. But if we set this transaction cost too high by allowing too much taking, we run the risk of eliminating the economic incentive for the creation of original works that is at the core of *1258copyright and — by driving creators out of the market — killing the proverbial goose that laid the golden egg.
Thus, the proper scope of the fair use doctrine in a given case boils down to an evidentiary question. As a conceptual matter, in making fair use determinations, we must conjure up a hypothetical perfect market for the work in question, consisting of the whole universe of those who might buy it, in which everyone involved has perfect knowledge of the value of the work to its author and to potential buyers, and excluding for the moment any potential fair uses of the work. Then, keeping in mind the purposes animating copyright law — the fostering of learning and the creation of new works — we must determine how much of that value the implied licensee-fair users can capture before the value of the remaining market is so diminished that it no longer makes economic sense for the author — or a subsequent holder of the copyright — to propagate the work in the first place.
In most instances, licensors (authors and copyright holders) and licensees (both paying licensees, and implied-by-law fair use licensees) will independently perform some version of this analysis in order to reach a mutually equitable arrangement. Ideally, a copyright holder will sell his or her works to buyers who pay the price that the market will bear and will routinely tolerate secondary uses which do not adversely impact that market. However, in the event of a disagreement, the copyright holder can file an infringement suit and the secondary user may invoke the fair use defense. In so doing, the parties essentially turn to a court to make a determination for them as to the appropriate boundaries of the secondary user’s implied license.
The fair use doctrine, as codified by Congress,19 furnishes judges with a laboratory within which to work to answer this question. The fair use statute provides:
Notwithstanding the provisions of 106 and 106A, the fair use of a copyrighted work, including such use by reproduction in copies or phonorecords or by any other means specified by that section, for purposes such as criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research, is not an infringement of copyright. In determining whether the use made of a work in any particular case is a fair use the factors to be considered shall include—
(1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
(2) the nature of the copyrighted work;
(3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
(4) the effect of the use upon the potential market for or value of the copyrighted work.
The fact that a work is unpublished shall not itself bar a finding of fair use if such finding is made upon consideration of all the above factors.
17 U.S.C. § 107.
In drafting § 107, Congress “resisted pressures from special interest groups to create presumptive categories of fair use, but structured the provision as an affirmative defense requiring a case-by-case analysis.”20 Harper & Row, 471 U.S. *1259at 561, 105 S.Ct. at 2231 (citing H.R.Rep. No. 83, 90th Cong., 1st Sess., 37 (1967)). Accordingly, the fair use inquiry is a flexible one. The four statutory factors provide courts with tools to determine— through a weighing of the four factors in light of the facts of a given case — whether a finding of fair use is warranted in that particular instance.
Thus, the examples enumerated in the preamble of § 107 — “criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research” — are meant to “give some idea of the sort of activities the courts might regard as fair use under the circumstances. This listing was not intended to be exhaustive, or to single out any particular use as presumptively a ‘fair’ use.” Id. (citations omitted) (quotation marks omitted); see also Peter Letterese & Assocs., Inc. v. World Inst. of Scientology Enters. Int’l, 533 F.3d 1287, 1308 (11th Cir.2008) (“The fair use doctrine is an ‘equitable rule of reason’; neither the examples of possible fair uses nor the four statutory factors are to be considered exclusive.” (quoting Abend, 495 U.S. at 236-37, 110 S.Ct. at 1768)). Furthermore, because fair use is an affirmative defense, its proponent bears the burden of proof in demonstrating that it applies. See Campbell, 510 U.S. at 590, 114 S.Ct. at 1177; Harper & Row, 471 U.S. at 561, 105 S.Ct. at 2231.
C.
Before we turn to the District Court’s analysis of each of the four fair use factors, we must first address the District Court’s overarching fair use methodology. Plaintiffs make two broad arguments that the District Court’s methodology was flawed, only one of which is persuasive.
Plaintiffs contend that the District Court erred by performing a work-by-work analysis that focused on whether the use of each individual work was fair use rather than on the broader context of ongoing practices at GSU. We disagree. Fair use must be determined on a case-by-case basis, by applying the four factors to each work at issue. See Campbell, 510 U.S. at 577, 114 S.Ct. at 1170. Were we to accept Plaintiffs’ argument, the District Court would have no principled method of determining whether a nebulous cloud of infringements purportedly caused by GSU’s “ongoing practices” should be excused by the defense of fair use. Thus, we find that the District Court’s work-by-work ap*1260proach — in which the District Court considered whether the fair use defense excused a representative sample of instances of alleged infringement in order to determine the need for injunctive relief — was the proper one.
Plaintiffs also argue that the District Court erred in giving each of the four factors equal weight, essentially taking a mechanical “add up the factors” approach, finding fair use if three factors weighed in favor of fair use and one against and vice versa, and only performing further analysis in case of a “tie.” We agree that the District Court’s arithmetic approach was improper.
Congress, in the Copyright Act, spoke neither to the relative weight courts should attach to each of the four factors nor to precisely how the factors ought to be balanced. However, the Supreme Court has explained that “the four statutory factors [may not] be treated in isolation, one from another. All are to be explored, and the results weighed together, in light of the purposes of copyright.” Id. at 578, 114 S.Ct. at 1170-71. In keeping with this approach, a given factor may be more or less important in determining whether a particular use should be considered fair under the specific circumstances of the case. See id. at 586, 114 S.Ct. at 1175 (noting that the second factor is generally not important in determining whether a finding of fair use is justified in the case of a parody). As such, the four factors “do not mechanistically resolve fair use issues.” Harper & Row, 471 U.S. at 588, 105 S.Ct. at 2245 (Brennan, J., dissenting). “Because [fair use] is not a mechanical determination, a party need not ‘shut-out’ her opponent on the four factor tally to prevail.” Wright v. Warner Books, Inc., 953 F.2d 731, 740 (2d Cir.1991). Accordingly, we find that the District Court erred in giving each of the four factors equal weight, and in treating the four factors as a simple mathematical formula. As we will explain, because of the circumstances of this case, some of the factors weigh more heavily on the fair use determination than others.
D.
We now turn to the District Court’s analysis of each individual fair use factor. Although we have found that the District Court’s method for weighing the four factors against one another was erroneous,this does not mean that the District Court’s reasoning under each of the four factors is also necessarily flawed. Rather, we must determine the correct analysis under each factor and then ascertain whether the District Court properly applied that analysis.
Plaintiffs argue that the District Court erred in its application of each of the four fair use factors. Plaintiffs’ argument centers on a comparison of the circumstances of the instant case to those of the so-called “eoursepack cases,” in which courts rejected a defense of fair use for commercial copyshops that assembled paper course-packs containing unlicensed excerpts of copyrighted works for use in university courses.
In Basic Books, Inc. v. Kinko’s Graphics Corp., publishing houses sued Kinko’s, a commercial copyshop, alleging that Kinko’s infringed the publishers’ copyrights when it copied excerpts from the publishers’ books, without permission and without payment of a license fee, and sold the copies for profit in bound, paper course-packs to students for use in college courses. 758 F.Supp. 1522, 1526 (S.D.N.Y.1991). The District Court rejected Kinko’s claim that its use of the excerpts was fair use, and granted injunctive relief to the publishers. Id.
*1261Similarly, in Princeton University Press v. Michigan Document Services, Inc., the Sixth Circuit upheld the District Court’s ruling that Michigan Document Services, a commercial copyshop, was not entitled to a fair use defense when it reproduced substantial portions of copyrighted academic works and sold the copies in bound, paper coursepacks to students for use in courses at the University of Michigan, without obtaining permission from the copyright holder. 99 F.3d 1381, 1383 (6th Cir.1996) (en banc). The Sixth Circuit held that injunctive relief was therefore warranted. Id. at 1392.
In essence, Plaintiffs argue that the coursepack cases should have guided the District Court’s analysis in this case, because GSU cannot alter the fair use calculus simply by choosing to distribute course readings in an electronic rather than paper format. In making this argument, Plaintiffs invoke the “media neutrality” principle, which “mandates that the ‘transfer of a work between media does not alter the character of that work for copyright purposes.’ ” See Greenberg v. Nat’l Geographic Soc., 533 F.3d 1244, 1257 (11th Cir.2008) (en banc) (quoting N.Y. Times Co., Inc. v. Tasini, 533 U.S. 483, 502, 121 S.Ct. 2381, 2392,150 L.Ed.2d 500 (2001)).
Plaintiffs’ reliance on the media neutrality doctrine is misplaced. Congress established that doctrine to ensure that works created with new technologies, perhaps not in existence at the time of the Copyright Act of 1976, would qualify for copyright protection. See id. (citing 17 U.S.C. § 102(a) (“Copyright protection subsists ... in original works of authorship fixed in any tangible medium of expression, now known or later developed....” (emphasis added))); see also H.R.Rep. No. 94-1476, at 52 (1976), reprinted in 1976 U.S.C.C.A.N. 5659, 5665 (“This broad language is intended to avoid the artificial and largely unjustifiable distinctions ... under which statutory copyrightability in certain cases has been made to depend upon the form or medium in which the work is fixed.”). The media neutrality doctrine concerns copyrightability and does not dictate the result in a fair use inquiry. Congress would not have intended this doctrine to effectively displace the flexible work-by-work fair use analysis in favor of a one dimensional analysis as to whether the case involves a transfer of a work between media.
Likewise, because the fair use analysis is highly fact-specific and must be performed on a work-by-work basis, see Cariou, 714 F.3d at 694, the coursepack cases provide guidance but do not dictate the results here, which must be based upon a careful consideration of the circumstances of the individual instances of alleged infringement involved in this case.
1.
The first fair use factor is “the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes.” 17 U.S.C. § 107(1). The inquiry under the first factor has several facets, including (1) the extent to which the use is a “transformative” rather than merely superseding use of the original work and (2) whether the use is for a nonprofit educational purpose, as opposed to a commercial purpose.' Peter Letterese & Assocs., 533 F.3d at 1309. “Before illumining these facets, however, we observe that the Supreme Court has cautioned against the use of the facets to create ‘hard evidentiary presumptionfs]’ or ‘categories of presumptively fair use.’ ” Id. (alteration in original) (citing Campbell, 510 U.S. at 584, 114 S.Ct. at 1174 (“[T]he mere fact that a use is educational and not for profit does not insulate it from a finding of infringement, *1262any more than the commercial character of a use bars a finding of fairness.”)).
Our initial inquiry under the first factor asks whether Defendants’ use is transformative, i.e., “whether the new work merely supersede^] the objects of the original creation, or instead adds something new, with a further purpose or different character, altering the first with new expression, meaning, or message.” Campbell, 510 U.S. at 579, 114 S.Ct. at 1171 (alteration in original) (citations omitted) (quotation marks omitted). For example, a parody transforms a work by appropriating elements of the work for purposes of comment or criticism, and thus “reflects transformative value because it ‘can provide social benefit, by shedding light on an earlier work, and, in the process, creating a new one.’ ” Suntrust Bank, 268 F.3d at 1271 (quoting Campbell, 510 U.S. at 579, 114 S.Ct. at 1171). A non-transformative use, on the other hand, is one which serves the same “overall function” as the original work. Peter Letterese & Assocs., 533 F.3d at 1311 (quotation marks omitted).
Even verbatim copying “may be transformative so long as the copy serves a different function than the original work.” Perfect 10, Inc. v. Amazon.com, Inc., 508 F.3d 1146, 1165 (9th Cir.2007) (finding a search engine’s copying of website images in order to create an Internet search index transformative because the original works “serve[d] an entertainment, aesthetic, or informative function, [whereas the] search engine transforms the image into a pointer directing a user to a source of information”); A.V. .v. iParadigms, LLC, 562 F.3d 630, 640 (4th Cir.2009) (finding use of student papers in an online plagiarism detection database transformative because the database used the papers not for their original purpose as schoolwork, but rather to automatically detect plagiarism in the works of other student authors); Bill Graham Archives v. Dorling Kindersley Ltd., 448 F.3d 605, 609 (2d Cir.2006) (finding use by publishers of concert posters reproduced in full, although in reduced size, in a biography of a musical group transforma-tive because the use was for historical and educational purposes, rather than advertising and informational purposes).
Allowing would-be fair users latitude for transformative uses furthers “the goal of copyright, to promote science and the arts.” See Campbell, 510 U.S. at 579, 114 S.Ct. at 1171. This is because transforma-tive works possess a comparatively large share of the novelty copyright seeks to foster. At the same time, transformative uses are less likely, generally speaking, to negatively impact the original creator’s bottom line, because they do not “ ‘merely supersede the objects of the original creation’ ” and therefore are less likely to “ ‘supplant’ the market for the copyrighted work [by] ‘fulfilling demand for the original.’ ” See Peter Letterese & Assocs., 533 F.3d at 1310 (alteration omitted) (quoting Campbell, 510 U.S. at 579, 588, 114 S.Ct. at 1171, 1176).
Here, Defendants’ use of excerpts of Plaintiffs’ works is not transformative. The excerpts of Plaintiffs’ works posted on GSU’s electronic reserve system are verbatim copies of portions of the original books which have merely been converted into a digital format. Although a professor may arrange these excerpts into a particular order or combination for use in a college course, this does not imbue the excerpts themselves with any more than a de minimis amount of new meaning. See Princeton University Press, 99 F.3d at 1389 (“[I]f you make verbatim copies of 95 pages of a 316-page book, you have not transformed the 95 pages very much— even if you juxtapose them to excerpts from other works.”).
*1263Nor do Defendants use the excerpts for anything other than the same intrinsic purpose — or at least one of the purposes— served by Plaintiffs’ works: reading material for students in university courses.21 Although an electronic reserve system may facilitate easy access to excerpts of Plaintiffs’ works, it does nothing to transform those works. But see Authors Guild, Inc. v. HathiTrust, 755 F.3d 87, 97 (2d Cir.2014) (holding that universities’ systematic digitization of copyrighted books was transformative because the digital copies were used to create a searchable database which supplied users with lists of page numbers and not with copies of the original works, and so the copies served a different purpose than the original works). Rather, Defendants’ use of excerpts of Plaintiffs’ works “supersede^] the objects of the original creation.” See Campbell, 510 U.S. at 579, 114 S.Ct. at 1171 (alteration in original) (quotation marks omitted). Were this element by itself dispositive, we would be compelled to find that the first factor weighs against a finding of fair use.
However, we must also consider under the first factor whether Defendants’ use is for a nonprofit educational purpose, as opposed to a commercial purpose. “[T]he commercial or non-transformative uses of a work are to be regarded as ‘separate factor[s] that tend[] to weigh against a finding of fair use,’ and ‘the force of that tendency will vary with the context.’ ” Peter Letterese & Assocs., 583 F.3d at 1309 (alteration in original) (quoting Campbell, 510 U.S. at 585, 114 S.Ct. at 1174). Indeed, the Supreme Court has recognized in dicta that nonprofit educational use may weigh in favor of a finding of fair use under the first factor, even when nontrans-formative. Campbell, 510 U.S. at 579 n. 11, 114 S.Ct. at 1171 n. 11 (“The obvious statutory exception to this focus on trans-formative uses is the straight reproduction of multiple copies for classroom distribution.”).
Because “copyright has always been used to promote learning,” Suntrust Bank, 268 F.3d at 1261, allowing some leeway for educational fair use furthers the purpose of copyright by providing students and teachers with a means to lawfully access works in order to further their learning in circumstances where it would be unreasonable to require permission. But, as al*1264ways, care must be taken not to allow too much educational use, lest we undermine the goals of copyright by enervating the incentive for authors to create the works upon which students and teachers depend.
In the eoursepack cases, Princeton University Press, 99 F.3d at 1389, and Basic Books, 758 F.Supp. at 1531-32, the first factor weighed against a finding of fair use when the nontransformative, educational use in question was performed by a for-profit copyshop, and was therefore commercial. In a more recent case, a district court refused to allow a commercial copyshop to sidestep the outcome of the eoursepack cases by requiring its student customers to perform the photocopying themselves (for a fee) when assembling paper coursepacks from master copies held by the copyshop. Blackwell Publ’g, Inc. v. Excel Research Grp., LLC, 661 F.Supp.2d 786, 794 (E.D.Mich.2009). In all three instances, the court refused to allow the defendants, who were engaged in commercial operations, to stand in the shoes of students and professors in claiming that their making of multiple copies of scholarly works was for nonprofit educational purposes.
However, in both of the eoursepack cases, the courts expressly declined to conclude that the copying would fall outside the boundaries of fair use if conducted by professors, students, or academic institutions. See Princeton University Press, 99 F.3d at 1389 (“As to the proposition that it would be fair use for the students or professors to make their own copies, the issue is by no means free from doubt. We need not decide this question, however, for the fact is that the copying complained of here was performed on a profit-making basis by a commercial enterprise.”); Basic Books, 758 F.Supp. at 1536 n. 13 (“Expressly, the decision of this court does not consider copying performed by students, libraries, nor on-campus copyshops, whether conducted for-profit or not.”). In Blackwell Publishing, the District Court noted that, conversely, “the fact that students do the copying does not ipso facto mean that a commercial use cannot be found.” 661 F.Supp.2d at 793.
Furthermore, where we previously held that the first factor weighed against a finding of fair use in a case involving use that was nontransformative but educational, the use in question was commercial. Peter Letterese & Assocs., 533 F.3d at 1309-12 (finding that the first factor weighed against a finding of fair use in a case involving the verbatim use of copyrighted material in an instructional course-pack for use by the Church of Scientology, where defendants charged a fee or obtained a promissory note in exchange for the coursepacks and hence the use was for commercial purposes).
Thus, the question becomes whether Defendants’ use of Plaintiffs’ works is truly a nonprofit educational use under § 107(1), and if so, whether this places sufficient weight on the first factor scales to justify a finding that this factor favors fair use despite the nontransformativeness of Defendants’ use.
GSU is a nonprofit educational institution. While this is relevant, our inquiry does not end there: we must consider not only the nature of the user, but the use itself. See Am. Geophysical Union v. Texaco Inc., 60 F.3d 913, 921-22 (2d Cir.1994) (“[A] court’s focus should be on the use of the copyrighted material and not simply on the user, [although] it is overly simplistic to suggest that the ‘purpose and character of the use’ can be fully discerned without considering the nature and objectives of the user.”).
Defendants’ use of Plaintiffs’ works in the teaching of university courses is clearly for educational purposes. Nevertheless, *1265it is not entirely clear that use by a nonprofit entity for educational purposes is always a “nonprofit” use as contemplated by § 107(1). The Supreme Court has explained that “[t]he crux of the profit/nonprofit distinction is not whether the sole motive of the use is monetary gain but whether the user stands to profit from exploitation of the copyrighted material without paying the customary price.” Harper & Row, 471 U.S. at 562, 105 S.Ct. at 2231. Plaintiffs point to several cases in which courts have found that educational use of copyrighted works by a nonprofit entity (or an individual associated with such an entity) was commercial even though the secondary user was not selling the items in question, in which “profit” took the form of an indirect economic benefit or a nonmonetary, professional benefit. See, e.g., Soc’y of Holy Transfiguration Monastery, Inc. v. Gregory, 689 F.3d 29, 61 (1st Cir.2012), cert. denied, — U.S. -, 133 S.Ct. 1315, 185 L.Ed.2d 195 (2013) (finding that the first factor weighed against fair use where an archbishop used copyrighted translations of a religious text on his website; although the use was educational, the archbishop profited from the use, in part, in the form of enhanced professional reputation); Worldwide Church of God v. Philadelphia Church of God, Inc., 227 F.3d 1110, 1118 (9th Cir.2000) (finding the first factor weighed against fair use where a religious organization distributed copies of a copyrighted book for use in its religious observance; the use was nontransformative, and although, the use was educational, the organization profited indirectly by using the work to attract new members who would tithe ten percent of their income); Weissmann v. Freeman, 868 F.2d 1313, 1324 (2d Cir.1989) (finding that the first factor weighed against fair use where a professor claimed an assistant’s paper as his own work and copied it for use in his class, under the professor’s name, because the professor profited from the use by enhancing his professional reputation and gaining a valuable authorship credit).
Under this line of reasoning, Defendants’ educational use of Plaintiffs’ works is a for-profit use despite GSU’s status as a nonprofit educational institution, and despite the fact that GSU does not directly sell access to Plaintiffs’ works on Eres and uLearn. Defendants “exploited” Plaintiffs’ copyrighted material for use in university courses without “paying - the customary price” — a licensing fee. Defendants profited from the use of excerpts of Plaintiffs’ works — however indirectly — because GSU collects money from students in the form of tuition and fees (which students pay in part for access to ERes and uLearn) and reduces its costs by avoiding fees it might have otherwise paid for the excerpts.
However, this reasoning is somewhat circular, and hence of limited usefulness to our fair use inquiry. Of course, any unlicensed use of copyrighted material profits the user in the sense that the user does not pay a potential licensing fee, allowing the user to keep his or her money. If this analysis were persuasive, no use could qualify as “nonprofit” under the first factor.' Moreover, if the use is a fair use, then the copyright owner is not entitled to charge for the use, and there is no “customary price” to be paid in the first place.22
*1266Accordingly, evaluating the indirect profit GSU gained by refusing to pay to license Plaintiffs’ works provides little useful guidance under the first factor. Simply put, the greater the amount of a work taken by the secondary user (or the more valuable the portion taken), the more the user “profits” by not paying for the use. Thus, the concern we have identified with profit in this sense is better dealt with under the third factor, which directs us to consider the amount of the original work that the secondary user appropriated, and the substantiality of the portion used. See 17 U.S.C. § 107(3).
Defendants’ use of Plaintiffs’ works does not provide GSU with a noneconomic but measurable professional benefit, such as an enhanced reputation. Contra Soc’y of Holy Transfiguration Monastery, 689 F.3d at 61. Although GSU students are likely pleased with the convenience of ERes and uLearn, there is no evidence that the presence of excerpts of Plaintiffs’ works on these electronic reserve systems enhances GSU’s reputation in any meaningful sense.23 There is no evidence that GSU gains any other measurable, indirect benefit by distributing Plaintiffs’ works to students, such as a valuable authorship credit either. Contra Weissmann, 868 F.2d at 1324.
Ultimately, we agree with the Second Circuit’s assessment that
[t]he commercial/nonprofit dichotomy concerns the unfairness that arises when a secondary user makes unauthorized use. of copyrighted material to capture significant revenues as a direct consequence of copying the original work. Consistent with these principles, courts will not sustain a claimed defense of fair use when the secondary use can fairly be characterized as a form of “commercial exploitation,” i.e., when the copier directly and exclusively acquires conspicuous financial rewards from its use of the copyrighted material. Conversely, courts are more willing to find a secondary use' fair when it produces a value that benefits the broader public interest. The greater the private economic rewards reaped by the secondary user (to the exclusion of broader public benefits), the more likely the first factor will favor the copyright holder and the less likely the use will be considered fair.
Texaco, 60 F.3d at 922 (2d Cir.1994) (citations omitted).24
*1267Although GSU certainly benefits from its use of Plaintiffs’ works by being able to provide the works conveniently to students, and profits in the sense that it avoids paying licensing fees, Defendants’ use is not fairly characterized as “commercial exploitation.” Even if Defendants’ use profits GSU in some sense, we are not convinced that this type of benefit is indicative of “commercial” use. There is no evidence that Defendants capture significant revenues as a direct consequence of copying Plaintiffs’ works. At the same time, the use 'provides a broader public benefit-furthering the education of students at a public university.
Thus, we find that Defendants’ use of Plaintiffs’ works is of the nonprofit educational nature that Congress intended the fair use defense to allow under certain circumstances. Furthermore, we find this sufficiently weighty that the first factor favors a finding of fair use despite the nontransformative nature of the use.
The text of the fair use statute highlights the importance Congress placed on educational use. The preamble to the statute provides that fair uses may include “teaching (including multiple copies for classroom use), scholarship, or research” and the first factor singles out “nonprofit educational purposes.” 17 U.S.C. § 107. The legislative history of § 107 further demonstrates that Congress singled out educational purposes for special consideration. In the years leading up to passage of the Copyright Act of 1976 (which introduced § 107), Congress devoted considerable attention to working out the proper scope of the fair use defense as applied to copying for educational and classroom purposes, going so far as to include in a final report the Classroom Guidelines developed by representatives of educator, author, and publisher groups at the urging of Congress. See H.R.Rep. No. 2237, at 59-66 (1966); S.Rep. No. 93-983, at 116-19 (1974); S.Rep. No. 94-473, at 63-65 (1975); H.R.Rep. No. 94-1476, at 66-70 (1976).
Notably, early drafts of § 107 did not include the parenthetical “including multiple copies for classroom use” or the specific direction to consider “whether [the] use is of a commercial nature or is for nonprofit educational purposes.” See S. 3008, H.R. 11947, H.R. 12354, 88th Cong. (1st Sess.1964); S. 1006, H.R. 4347, H.R. 5680, H.R. 6831, H.R. 6835, 89th Cong. (1st Sess. 1965); S. 597, H.R. 2512, H.R. 5650, 90th Cong. (1st Sess.1967). This language was not inserted until one month before the passage of the Copyright Act of 1976. See S. 22, 94th Cong. (2d Sess.1976).
In sum, Congress devoted extensive effort to ensure that fair use would allow for educational copying under the proper circumstances and wás sufficiently determined to achieve this goal that it amended the text of the statute at the eleventh hour in order to expressly state it. Furthermore, as described above, allowing latitude for educational fair use promotes the goals of copyright. Thus, we are .persuaded that, despite the recent focus on transfor-mativeness under the first factor, use for teaching purposes by a nonprofit, educational institution such as Defendants’ favors a finding of fair use under the first factor, despite the nontransformative nature of the use.
Accordingly, we find that the District Court did not err in holding that the first factor favors a finding of fair use. Nevertheless, because Defendants’ use of Plaintiffs’ works is nontransformative, the threat of market substitution is significant. We note that insofar as the first factor is concerned with uses that supplant demand for the original, this factor is “closely related” to “[t]he fourth fair use factor, the effect on the potential market for the work.” See Pac. & S. Co., 744 F.2d at *12681496. We will thus revisit this concern when we analyze the fourth factor. See infra part III.D.4.
2.
The second' fair use factor, “the nature of the copyrighted work,” 17 U.S.C. § 107(2), “calls for recognition that some works are closer to the core of intended copyright protection than others, with the consequence that fair use is more difficult to establish when the former works are copied,” Campbell, 510 U.S. at 586, 114 S.Ct. at 1175. The inquiry under the second factor generally focuses on two criteria. First, because works that are highly creative are closer to the core of copyright — that is, such works contain the most originality and inventiveness — the law affords such works maximal protection, and hence it is less likely that use of such works will be fair use.25 Id. In contrast, “[t]he law generally recognizes a greater need to disseminate factual works than works of fiction or fantasy,” and so it is more likely that the use of a factual or informational work will be fair use. Harper & Row, 471 U.S. at 563, 105 S.Ct. at 2232; see also Stewart v. Abend, 495 U.S. 207, 237, 110 S.Ct. 1750, 1769, 109 L.Ed.2d 184 (1990). Second, because an “author’s right to control the first public appearance of his expression weighs against such use of the work before its release,” use of an unpublished work is less likely to be fair use. Harper & Row, 471 U.S. at 564, 105 S.Ct. at 2232. Because all of Plaintiffs’ works in question here are published, we will focus on the creative/factual distinction.
A paradigmatic example of a creative work, the use of which will disfavor fair use under the second factor, is “[a] motion picture based on a fictional short story.” Abend, 495 U.S. at 238, 110 S.Ct. at 1769. On the factual end of the spectrum, secondary use of a “bare factual compilation ]” favors fair use under the second factor. Campbell, 510 U.S. at 586, 114 S.Ct. at 1175 (citing Feist, 499 U.S. at 348-51, 111 S.Ct. at 1289-91). However, “[e]ven within the field of fact[ual] works, there are gradations as to the relative proportion of fact and fancy. One may move from sparsely embellished maps and directories to elegantly written biography.” Harper & Row, 471 U.S. at 563, 105 S.Ct. at 2232 (quoting Robert A. Gorman, Fact or Fancy? The Implications for Copyright, 29 J. Copyright Soc’y 560, 561 (1982)).
The coursepack cases — which involved copying of academic works similar to those involved here — reached opposite conclusions as to the effect of the second factor.26 *1269Compare Princeton Univ. Press, 99 F.3d at 1389 (“[T]he excerpts copied for the coursepacks contained creative material, or ‘expression;’ it was certainly not telephone book listings that the defendants were reproducing. This-¡factor ... cuts against a finding of fair use.”), with Basic Books, 758 F.Supp. at 1533 (“The books infringed in suit were factual in nature. This factor weighs in favor of defendant.”).
Nevertheless, relevant precedent indicates the proper approach. In Harper & Row, a publisher holding exclusive rights to President Ford’s unpublished memoirs sued The Nation magazine after The Nation published portions of the memoirs. 471 U.S. at 543, 105 S.Ct. at 2221-22. Although it focused on the unpublished nature of the memoir, the Court held that the second factor disfavored fair use in part because “The Nation did not stop at isolated phrases and instead excerpted subjective descriptions and portraits of public figures whose power lies in the author’s individualized expression. Such use, focusing on the most expressive elements of the work, exceeds that necessary to disseminate the facts.” Id. at 563-64, 105 S.Ct. at 2232.
In Peter Letterese & Associates, the holder of the copyright on a book about sales techniques sued several entities associated with the Church of Scientology after the entities used portions of the book in materials prepared for Church training courses. 533 F.3d at 1294-96. We held that the second factor was neutral — i.e., did not weigh for or against fair use — in part because, although the book
[fell] roughly under the rubric of a factual work[,] ... [it] contained] a significant “proportion of fact and fancy,” and not merely in the subjective selection and arrangement of sales techniques; [the author] utilize[d] original expression that surpasses the bare facts necessary to communicate the underlying technique. Although the techniques are presented by way of personal anecdote, it is hard to believe that such anecdotes feature actual persons and actual retellings of past events and conversations, as opposed to composite characters and experiences served with a healthy dose of fiction.
Id. at 1312-13.
Similarly, in Marcus v. Rowley, the Ninth Circuit considered a case involving the unauthorized copying of portions of the plaintiffs instructional booklet on the subject of cake decoration. 695 F.2d 1171, 1172 (9th Cir.1983). The court held that
plaintiffs booklet involved both informational and creative aspects. Some pages in her booklet undoubtedly contained information available in other cake decorating books or in recipe books. Other parts of her booklet contained creative hints she derived from her own experiences or ideas; certainly the manner in which plaintiff assembled her book represented a creative expression. Thus, on balance, it does not appear that analysis of this factor is of any real assistance in reaching a conclusion as to applicability of fair use.
Id. at 1176.
Here, the District Court held that “[b]ecause all of the excerpts are informational and educational in nature and none are fictional, fair use factor two weighs in favor of Defendants.” Cambridge Univ. Press, 863 F.Supp.2d at 1242. We disagree.
The District Court found that “[s]ome of the books [at issue] are not merely descriptive; they contain material of an evaluative nature, giving the authors’ perspec*1270tives and opinions.” Id. at 1226. Such material might involve “subjective descriptions [that rely on] the author’s individualized expression,” Harper & Row, 471 U.S. at 563, 105 S.Ct. at 2232, may “surpass[ ] the bare facts necessary to communicate the underlying” information, Peter Letterese & Assocs., 533 F.3d at 1312, or may be “derived from [the author’s] own experiences,” Rowley, 695 F.2d at 1176. Although there appears to be no evidence that any of the non-fiction works in question here are “served with a healthy dose of fiction,” neither are all of the works mere “factual compilation[s].”27 Peter Letterese & Assocs., 533 F.3d at 1312-13.
Defendants argue that GSU professors chose the excerpts of Plaintiffs’ works for their factual content, not for any expressive content the works may contain, noting that several professors testified that if the use of a particular excerpt was not a fair use, they would have found another source. Of course, other professors testified that they chose particular excerpts because of the author’s interpretative originality and significance. Regardless of whether GSU faculty chose the excerpts for their expressive or factual content, the excerpts were copied wholesale — facts, ideas, and original expression alike. Which aspect the secondary user was interested in is irrelevant to the disposition of the second factor.
Accordingly, we find that the District Court erred in holding that the second factor favored fair use in every instance. Where the excerpts of Plaintiffs’ works contained evaluative, analytical, or subjectively descriptive material that surpasses the bare facts necessary to communicate information, or derives from the author’s experiences or opinions, the District Court should have held that the second factor was neutral, or even weighed against fan-use in cases of excerpts that were dominated by such material. That being said, the second fair use factor is of relatively little importance in this case.28
*12713.
The third fair use factor is “the amount and substantiality of the portion used in relation to the copyrighted work as a whole.” 17 U.S.C. § 107(3). “[T]his third factor examines whether defendants have ‘helped themselves overmuch’ of the copyrighted work in light of the purpose and character of the use.” Peter Letterese & Assocs., 533 F.3d at 1314 (quoting Campbell, 510 U.S. at 587, 114 S.Ct. at 1175). Thus, this factor is intertwined with the first factor.
“[T]his factor is [also] intertwined with the fourth factor and partly functions as a heuristic to determine the impact on the market for the original.” Id. (footnote omitted). As we have explained,
A book reviewer who copies snippets of a book is likely to increase the demand for the book, but “were a book reviewer to quote the entire book in his review, or so much of the book as to make the review a substitute for the book itself, he would be cutting into the publisher’s market, and the defense of fair use would fail.” Id. (quoting Ty, Inc. v. Publ’ns Int'l Ltd., 292 F.3d 512, 517 (7th Cir.2002)). Thus, “[t]he inquiry is whether the amount taken is reasonable in light of the purpose of the use and the likelihood of market substitution.” Id. at 1314 n. 30 (citing Campbell, 510 U.S. at 588, 114 S.Ct. at 1176).
In making this determination, we must consider “not only ... the quantity of the materials used, but ... their quality and importance, too.” Campbell, 510 U.S. at 587, 114 S.Ct. at 1175; see also Harper & Row, 471 U.S. at 565, 105 S.Ct. at 2233 (holding that the third factor disfavored fair use because the defendant copied a qualitatively substantial portion of the original work — “the most interesting and moving parts of the entire manuscript” or “the heart of the book” — even though the defendants copied only approximately 300 words out of the 200,000 words in the plaintiffs’ work (quotation marks omitted)).
Here, the District Court found that the third factor favored fair use in instances where Defendants copied no more than 10 percent of a work, or one chapter in case of a book with ten or more chapters. Cambridge Univ. Press, 863 F.Supp.2d at 1243. The District Court’s blanket 10 percent-or-one-chapter benchmark was improper. The fair use analysis *1272must be performed on a case-by-case/work-by-work basis. Campbell, 510 U.S. at 577, 114 S.Ct. at 1170; see supra discussion accompanying note 20. We must avoid “hard evidentiary presumption[s] ... and ‘eschew[ ] a rigid, bright-line approach to fair use.’ ” Campbell, 510 U.S. at 584-85, 114 S.Ct. at 1174 (quoting Sony, 464 U.S. at 449 n. 31, 104 S.Ct. at 792 n. 31). By holding that the third factor favored fair use whenever the amount of copying fell within a 10 percent-or-one-chapter baseline, the District Court abdicated its duty to analyze the third factor for each instance of alleged infringement individually.
Defendants argue that the District Court’s 10 percent-or-one-chapter baseline served as a starting point only. However, this “starting point” in fact served as a substantive safe harbor in the third factor analysis, an approach which is incompatible with the prescribed work-by-work analysis. Even if we consider the baseline as a starting point only, application of the same non-statutory starting point to each instance of infringement is not a feature of a proper work-by-work analysis under the third fair use factor.
Defendants also argue that the District Court’s 10 percent-or-one-chapter approach is supported by the record. Defendants’ explain that a CCC white paper, Using Electronic Reserves: Guidelines and Best Practices for Copyright Compliance (2011), identifies “best practices” for electronic reserves, stating that electronic reserve materials should be limited to “small excerpts” and that “[m]ost experts advise using a single article or ... chapter of a copyrighted work....” See Defendants’ Trial Ex. 906, at 2.- However, even if we accept that the 10 percent-or-one-chapter approach represents a general industry “best practice” for electronic reserves, this is not relevant to an individualized fair use analysis.
Plaintiffs offer four additional critiques of the District Court’s analysis under the third factor. First, Plaintiffs argue that the District Court erred in focusing its inquiry on whether the amount copied suited GSU’s pedagogical purposes. This analysis, Plaintiffs contend, reflects a misplaced reliance on Campbell, in which the Supreme Court held that it was proper to allow a parodist — a transformative user— to use “at least enough” of the original work “to make the object of its critical wit recognizable.” 510 U.S. at 588, 114 S.Ct. at 1176; see also Suntrust Bank, 268 F.3d at 1271. Such an analysis, Plaintiffs conclude, has no application in a ease of non-transformative copying like this one.
Plaintiffs mischaracterize the holding of Campbell. Although Campbell involved a parody, the Supreme Court’s statement in that case that the inquiry under the third factor is whether “the quantity and value of the materials used ... are reasonable in relation to the purpose of the copying” was not limited to the context of transformative uses. See 510 U.S. at 586, 114 S.Ct. at 1175 (citations omitted) (quotation marks omitted). Indeed, the Supreme Court acknowledged in Campbell that it had recognized that “the extent of permissible copying varies with the purpose and character of the use” in prior cases that did not involve transformative uses. Id. at 586-87, 114 S.Ct. at 1175 (citing Sony, 464 U.S. at 449-50, 104 S.Ct. at 792-93 (holding that reproduction of an entire work “does not have its ordinary effect of militating against a finding of fair use” as to home videotaping of television programs); Harper & Row, 471 U.S. at 564, 105 S.Ct. at 2232 (“[E]ven substantial quotations might qualify as fair use in a review of a published work or a news account of a speech” but not in a scoop of a soon-to-be-published memoir)). We have also previously held that this principle applies in the con*1273text of nontransformative uses. Peter Letterese & Assocs., 538 F.3d at 1314-15. Thus, we find that the District Court properly took into account whether the amount copied suited GSU’s pedagogical purposes.
Second, Plaintiffs argue that the District Court erred in measuring the amount taken based on the length of the entire book even where the copied material was an independently authored chapter in an edited volume. Rather, Plaintiffs contend, the relevant “work” in the case of an edited volume is the chapter copied, not the entire book; to conclude otherwise would create the anomalous result that a work bound with other works in an edited volume would enjoy less copyright protection than if the same work were published in a journal. See Texaco, 60 F.3d at 926 (treating individual articles in a journal as discrete works of authorship for purposes of third factor analysis).
As noted earlier, the District Court declined to consider this argument because Plaintiffs raised it late in the proceedings. The decision whether to hear an argument raised late in litigation is squarely within the discretion of the District Court. See Foman v. Davis, 371 U.S. 178, 182, 83 S.Ct. 227, 230, 9 L.Ed.2d 222 (1962). In the parties’ March 15, 2011, joint exhibit — the document, which was admitted into evidence, detailing the alleged instances of infringement which were to be the subject of the bench trial — Plaintiffs identified whole books as the subject of the infringement claims and calculated the percentages taken as the amount copied from the total number of pages in those books. In their pre-trial proposed Findings of Fact and Conclusions of Law, filed on May 17, 2011, Plaintiffs again calculated the percentages in this manner. Although the argument that individual chapters should constitute the whole work in edited volumes for the purposes of factor three “came up for the first time briefly during the trial, ... [i]t was not until Plaintiffs filed their post-trial Proposed Findings of Fact and Conclusions of Law that Plaintiffs’ ... theory was fleshed out.” Cambridge Univ. Press, 863 F.Supp.2d at 1231.
Plaintiffs argue that Defendants had an opportunity to respond to the argument in their responsé to Plaintiffs’ post-trial Proposed Findings of Fact and Conclusions of Law, and that the District Court identified no prejudice to Defendants. However, the bench trial was conducted based on the pre-trial filings, in which Plaintiffs’ calculated the percentages in relation to whole works. We find that the District Court did not abuse its discretion in rejecting Plaintiffs’ argument regarding chapters of edited compilations “on grounds of untimeliness and unfair surprise to the Defendants.” Id.
Third, Plaintiffs argue that the copying permitted by the District Court exceeds the amounts outlined in the Classroom Guidelines.29 We note that the Classroom Guidelines, although part of the legislative history of the Copyright Act, do not carry force of law. In any case, to treat the Classroom Guidelines as indicative of what is allowable would be to create the type of “hard evidentiary presumption” that the Supreme Court has cautioned against, because fair use must operate as a “ ‘sensitive balancing of interests.’ ” Campbell, 510 U.S. at 584, 114 S.Ct. at 1174 (quoting Sony, 464 U.S. at 455, n. 40, 104 S.Ct. at 795, n. 40). As discussed, the fair use analysis must be performed on a work-by-*1274work basis, and so we must not give undue weight to the amounts of copying set forth in the Classroom Guidelines.
Furthermore, although Plaintiffs characterize the amounts set forth in the Classroom Guidelines as “limits,” the Classroom Guidelines were intended to suggest a minimum, not maximum, amount of allowable educational copying that might be fair use, and were not intended to limit fair use in any way:
The purpose of the [Classroom] [Guidelines is to state the minimum and not the maximum standards of educational fair use....
Moreover, the following statement of guidelines is not intended to limit the types of copying permitted under the standards of fair use under judicial decision and which are stated in Section 107 of the Copyright Revision Bill.
H.R.Rep. No. 94-1476, at 68 (1976), reprinted in U.S.C.C.A.N. 5659, 5681. Thus, while the Classroom Guidelines may be seen to represent Congress’ tentative view of the permissible amount of educational copying in 1976, we are not persuaded by the Plaintiffs’ argument that the Classroom Guidelines should control the analysis under factor three in this case.
Finally, Plaintiffs argue that the District Court allowed excessive taking as a percentage of the entire book compared to the amounts held to be “over the line” in the coursepack cases. The District Court erred, Plaintiffs contend, in finding that because GSU’s use was nonprofit and educational, “fair use factor one strongly favors Defendants and tends to push the amount of permissible copying toward a greater amount than” the amounts held to disfavor fair use in the coursepack cases. Cambridge Univ. Press, 863 F.Supp.2d at 1232. In Basic Books, the court found the fact that certain excerpts represented 5 to 14 percent of the whole work to weigh against the defendant, and the fact that other excerpts represented 16 to 28 percent of the whole work to “weigh heavily against [the] defendant.” 758 F.Supp. at 1527-28. In Princeton University Press, the court found that takings ranging from 5 to 30 percent of the works in question weighed against fair use. 99 F.3d at 1391. Plaintiffs point out that the portions of Plaintiffs’ works that the District Court held to favor fair use under the third factor are comparable to those held to disfavor fair use in the coursepack cases.
We first note that the coursepack cases are not binding authority on this Court. Furthermore, because the four factors must be “weighed together” and not “treated in isolation,” Campbell, 510 U.S. at 578, 114 S.Ct. at 1171, it is appropriate for the District Court to take the educational purpose of the use into consideration when analyzing how much copying is permissible under the third factor. This must be done on a work-by-work basis in a case such as this. While this type of analysis necessarily precludes hard-and-fast evidentiary presumptions, the wholesale reproduction of an entire work will not generally be considered fair unless the use is highly transformative.30
*1275Accordingly, we find that the District Court properly considered whether the individual instances of alleged infringement were excessive in relation to Defendants’ pedagogical purpose, properly measured the amounts taken in all cases based on the length of the entire book, and properly declined to tie its analysis under the third factor to the Classroom Guidelines or to the coursepack cases. However, we find that the District Court erred in applying a 10 percent-or-one-chapter safe harbor in it analysis of the individual instances of alleged infringement. The District Court should have analyzed each instance of alleged copying individually, considering the quantity and the quality of the material taken — including whether the material taken constituted the heart of the work — and whether that taking was excessive in light of the educational purpose of the use and the threat of market substitution.
4.
The fourth fair use factor is “the effect of the use upon the potential market for or value of the copyrighted work.” 17 U.S.C. § 107(4). “We must consider two inquiries: (1) ‘the extent of the market harm caused by the particular actions of the alleged infringer,’ and (2) ‘whether unrestricted and widespread conduct of the sort engaged in by the defendant ] would result in a substantially adverse impact on the potential market.’” Peter Letterese & Assocs., 533 F.3d at 1315 (alteration in original) (quoting Campbell, 510 U.S. at 590, 114 S.Ct. at 1177 (quotation marks omitted)). The adverse impact we are “primarily concerned [with] is that of market substitution.” Id.; see also Harper & Row, 471 U.S. at 568, 105 S.Ct. at 2235 (explaining that the fourth factor is concerned with “use that supplants any part of the normal market for a copyrighted work” (quoting S.Rep. No. 94-473, at 65 (1975))). Furthermore, “[mjarket harm is a matter of degree, and the importance of [the fourth] factor will vary, not only with the amount of harm, but also with the relative strength of the showing on the other factors.” Campbell, 510 U.S. at 590 n. 21, 114 S.Ct. at 1177 n. 21. Because Defendants’ use is nontransformative and fulfills the educational purposes that Plaintiffs, at least in part, market their works for, the threat of market substitution here is great and thus the fourth factor looms large in the overall fair use analysis.31
*1276The central question under the fourth factor is not whether Defendants’ use of Plaintiffs’ works caused Plaintiffs to lose some potential revenue. Rather, it is whether Defendants’ use-taking into account the damage that might occur if “everybody did it”-would cause substantial economic harm such that allowing it would frustrate the purposes of copyright by materially impairing Defendants’ incentive to publish the work. See Harper & Row, 471 U.S. at. 566-67, 105 S.Ct. at 2234 (“Fan-use, when properly applied, is limited to copying by others which does not materially impair the marketability of the work which is copied.” (emphasis added) (quotation marks omitted)).
We agree with the District Court that the small excerpts Defendants used do not substitute for the full books from which they were drawn. “Plaintiffs offered no trial testimony or evidence showing that they lost any book sales in or after 2009 on account of any actions by anyone at Georgia State.” Cambridge Univ. Press, 863 F.Supp.2d at 1217. Thus, the District Court did not err in finding that “Defendants’ use of small excerpts did not affect Plaintiffs’ actual or potential sales of books.” Id. at 1236.
However, CCC’s various programs for academic permissions-and Plaintiffs’ own permissions programs-constitute a workable market through which universities like GSU may purchase licenses to use excerpts of Plaintiffs’ works. Plaintiffs contend that, by failing to purchase digital permissions to use excerpts of Plaintiffs’ works on ERes and uLearn, Defendants caused substantial harm to the market for licenses, and that widespread adoption of this practice would cause substantial harm to the potential market. Plaintiffs also argue that, even if a license for a digital excerpt of a work was unavailable, this should not weigh in favor of fair use because the copyright owner is not obliged to accommodate prospective users.
Defendants argue that, because permissions income for academic books represents a miniscule percentage of Plaintiffs’ overall revenue, Defendants’ practices have not caused substantial harm to the market for Plaintiffs works, and would not do so even if widely adopted. Defendants further argue that unavailability of licensing opportunities for particular works should weigh in favor of fair use.
We note that it is not determinative that programs exist through which universities may license excerpts of Plaintiffs’ works. In other words, the fact that Plaintiffs have made paying easier does not automatically dictate a right to payment. “[A] copyright holder can always assert some degree of adverse [effect] on its potential licensing revenues as a consequence of the secondary use at issue simply because the copyright holder has not been paid a fee to permit that particular use.” Texaco, 60 F.3d at 929 n. 17 (citations omitted). The goal of copyright is to stimulate the creation of new works, not to furnish copyright holders with control over all markets. Accordingly, the ability to license does not demand a finding against fair use.
Nevertheless, “it is sensible that a particular unauthorized use should be considered ‘more fair’ when there is no ready market or means to pay for the use, while such an unauthorized use should be considered ‘less' fair’ when there is a *1277ready market or means to pay for the use. The vice of circular reasoning arises only if the availability of payment is conclusive against fair use.” Id. at 981. Put simply, absent evidence to the contrary, if a copyright holder has not made a license available to use a particular work in a particular manner, the inference is that the author or publisher did not think that there would be enough such use to bother making a license available. In such a case, there is little damage to the publisher’s market when someone makes use of the work in that way without obtaining a license, and hence the fourth factor should generally weigh in favor of fair use. This is true of Plaintiffs’ works for which no license for a digital excerpt was available.32
Plaintiffs argue that even though a use is less fair when licensing is readily available, it does not follow that a use becomes more fair if, for a legitimate reason, the copyright holder has not offered to license the work. Plaintiffs cite several cases which have found that the fourth factor weighs against fair use even though the copyright holder was not actively marketing the work in question because the secondary use negatively impacted the potential market for the work. See Balsley v. LFP, Inc., 691 F.3d 747, 761 (6th Cir. 2012), cert. denied, — U.S. -, 133 S.Ct. 944, 184 L.Ed.2d 727 (2013) (finding that a magazine publisher had failed to rebut a presumption of market harm in a commercial context where the publisher published a photo in which plaintiffs had acquired copyright in order to prevent its dissemination, because plaintiffs’ “current desire or ability to avail themselves of the market” was irrelevant to the question of potential market harm); Worldwide Church of God, 227 F.3d at 1119 (finding market harm even though the copyright holder had ceased distributing the copyrighted work ten years prior to the act of alleged infringement and had no concrete plans to publish a new version); Castle Rock Entm’t, Inc. v. Carol Publ’g Grp., Inc., 150 F.3d 132, 145-46 (2d Cir.1998) (finding that the fourth factor weighed against a defendant who published a book containing trivia questions about the plaintiffs copyrighted television program where the plaintiff “ha[d] evidenced little if any interest in exploiting [the] market for derivative works based on” the program, noting that “copyright law must respect that creative and economic choice”). We note that our own precedent also supports this theory in some circumstances. See Pac. & S. Co., 744 F.2d at 1496 (finding harm to the potential market for plaintiffs news broadcasts where a defendant videotaped the broadcasts and sold tapes to the subjects of the news reports because “[c]opy-rights protect owners who immediately market a work no more stringently than owners who delay before entering the market” and so “[t]he fact that [the plaintiff] does not actively market copies of the news programs does not matter, for Section 107 looks to the ‘potential market’ in analyzing the effects of an alleged infringement”).
However, this reasoning need not dictate the result in this case, which concerns not the market for Plaintiffs’ original works themselves or for derivative works based upon those works, but rather a mar*1278ket for licenses to use Plaintiffs’ works in a particular way. As previously explained, licensing poses a particular threat that the fair use analysis will become circular, and Plaintiffs may not head off a defense of fair use by complaining that every potential licensing opportunity represents a potential market for purposes of the fourth fair use factor.
An analogy is helpful. A publisher acts like a securities underwriter. A publisher determines the value of a work, which is set by the anticipated demand for the work. Thus, the greater the demand for the work — the greater the market — the more the publisher will pay the author of the work up front, and the more the publisher will endeavor to make the work widely available. If a publisher makes licenses available for some uses but not for others, this indicates that the publisher has likely made a reasoned decision not to enter the licensing market for those uses, which implies that the value of that market is minimal.
With regard to the works for which digital permissions were unavailable, Plaintiffs choose to .enter those works into some markets — print copies of the whole work, or perhaps licenses for paper copies of excerpts — but not the digital permission market. This tells us that Plaintiffs likely anticipated that there would be little to no demand for digital excerpts of the excluded works and thus saw the value of that market as de minimis or zero. If the market for digital excerpts were in fact de minimis or zero, then neither Defendants’ particular use nor a widespread use of similar kind would be likely to cause significant market harm. Of course, if publishers choose to participate in the market the calculation will change.
In its individual analysis under the fourth factor of each of the forty-eight works for which it found Plaintiffs had made a prima facie case of infringement, the District Court performed a sufficiently nuanced review of the evidence regarding license availability. Where the evidence showed that there was a ready market for digital excerpts of a work in 2009, the time of the purported infringements, the District Court found that there was small— due to the amount of money involved — but actual damage to the value of Plaintiffs’ copyright.33 The District Court also properly took into account that widespread use of similar unlicensed excerpts could cause substantial harm to the potential market. Thus, where there was a license for digital excerpts available, the District Court generally held that the fourth factor weighed against a finding of fair use. In close cases, the District Court went further and *1279examined the amount of permissions income a work had generated in order to determine how much this particular revenue source contributed to the value of the copyright in the work, noting that where there is no significant demand for excerpts, the likelihood of repetitive unpaid use is diminished. Where there was no evidence in the record to show that a license for digital excerpts was available— as was the case for seventeen works published by Oxford and Cambridge — the District Court held that the fourth factor weighted in favor of fair use. We find that the District Court’s analysis under the fourth factor was correct, and that the District Court properly took license availability into account in determining whether the fourth factor weighted for or against fair use.
Plaintiffs argue that the District Court erred by placing the burden on Plaintiffs to show that digital licenses for the particular works in question were reasonably available through CCC in 2009. Cognizant that fair use is an affirmative defense, the District Court kept the overall burden on Defendants to show that “no substantial damage was caused to the potential market for or the value of Plaintiffs’ works” in order to prevail on the question of whether the fourth factor should favor fair use. Cambridge Univ. Press, 863 F.Supp.2d at 1237. However, the District Court found that because Plaintiffs were “advocates of the theory that the availability of licenses shifts the factor four fair use analysis in their favor ... it is appropriate for them to be called upon to show that CCC provided in 2009 reasonably efficient, reasonably priced, convenient access to the particular excerpts which are in question in this case.” Id. Plaintiffs argue that this amounted to relieving the Defendants of their burden of proof on the fourth factor.
We disagree. Fair use is an affirmative defense, and the evidentiary burden on all four of its factors rests on the alleged infringer. See Campbell, 510 U.S. at 590, 114 S.Ct. at 1177. However, Plaintiffs—as publishers — can reasonably be expected to have the evidence as to availability of licenses for their own works. It is therefore reasonable to place on Plaintiffs the burden of going forward with the evidence on this question.
In effect, this creates a presumption that no market for digital permissions exists for a particular work.34 Cf. Verbraeken v. Westinghouse Elec. Corp., 881 F.2d 1041, 1045 (11th Cir.1989) (explaining that, in a ease brought pursuant to the Age Discrimination in Employment Act, “the plaintiff bears the ultimate burden of proving that age was a determinative factor in the employer’s decision to terminate his employment,” but that after the plaintiff establishes a prima facie case of discrimination, this creates a presumption that age was a determinative factor and the defendant then has the burden to go forward with evidence of “a legitimate, non-discriminatory rationale for the discharge,” after which the plaintiff must prove that the employer’s purported reason was merely a pretext). This is reasonable, because if a license was available during the relevant time period, Plaintiffs can rebut the presumption of no market by going *1280forward with evidence of license availability. If there is evidence of a potential, future market, Plaintiffs can rebut the presumption by going forward with that. Then, Defendants — retaining the overall burden of persuasion on the fourth factor — must demonstrate that their use does not materially impair the existing or potential market in order to prevail.
Although the District Court did not articulate its approach to the evidentiary burden on license availability in exactly this manner, the District Court did essentially what we have described. The District Court required Plaintiffs to put on evidence as to the availability of digital permissions in 2009, and Plaintiffs provided such evidence for some of the works in question but not for others. For those seventeen works for which Plaintiffs presented no evidence that digital permissions were available, the District Court — noting that, because access was limited to particular classes, it was unlikely that Defendants’ use would result in exposure of the works to the general public and so there was little risk of widespread market substitution for excerpts of the works — held that there was no harm to the actual or potential market. For those works for which Plaintiffs demonstrated that digital permissions were available, the District Court considered the evidence demonstrating that the actual harm to the value of Plaintiffs’ copyright was minor (because the fees Defendants would have paid for a small number of licenses for the works in question amounted to a relatively small amount), but reasonably concluded that widespread conduct of the sort engaged in by Defendants would cause substantial harm.35 Thus, although the District Court required Plaintiffs to go forward with evidence of license availability, the District Court properly kept the ultimate burden of persuasion on Defendants on the question of market harm under the fourth factor. Accordingly, the District Court did not engage in improper burden shifting.36
*1281Finally, Plaintiffs argue that the District Court erred in finding that digital licenses for many of the works in question were unavailable in 2009. Plaintiffs point to the joint exhibit enumerating the allegations of infringement which were to be the basis of the trial, filed on March 15, 2011. Plaintiffs note that this document identified what it would have cost in 2009 to license each of the works at issue through CCC. Defendants, however, argue that the joint exhibit merely depicts what the permission fee for each work might have been using the standard per-page rates charged by CCC for each publisher. The District Court accepted this argument, noting that “[t]he fact that a license to copy an excerpt of an individual work would have cost a particular amount is not a substitute for evidence that the license was actually available.” Cambridge Univ. Press, 863 F.Supp.2d at 1238.
Given the evidence in the record, it was reasonable for the District Court to find that a lack of direct evidence of availability indicated that licenses were unavailable. Cambridge’s representative testified that Cambridge does not allow excerpts of certain categories of books to be licensed through CCC, including reference and language books (as several of the works at issue here are). Representatives of Oxford, Cambridge, and CCC testified, but did not explain whether digital licenses for excerpts were available in 2009 for the books at issue in this case.37 The District Court also noted .that the record shows that Cambridge and Oxford have been reluctant to make digital excerpts available. Thus, the District Court’s findings of fact regarding license availability are not clearly erroneous.
The District Court engaged in a careful investigation of the evidence in the record, properly considered the availability of digital permissions in 2009, and appropriately placed the burden of going forward with the evidence on this issue on Plaintiffs. Accordingly, we find that the District Court did not err in its application of the fourth factor. However, because Defendants’ copying was nontransformative and the threat of market substitution was therefore serious, the District Court erred by not affording the fourth factor additional weight in its overall fair use calculus.
5.
The District Court enumerated two additional, purportedly non-statutory considerations which it held favored fair use: (1) that “[l]imited unpaid copying of excerpts will not deter academic authors from creating new academic works,” and *1282(2) that “[t]he slight limitation of permissions income caused by the fair use authorized by this Order will not appreciably diminish Plaintiffs’ ability to publish scholarly works and will promote the spread of knowledge.” Id. at 1240. Although it is within the District Court’s discretion to go beyond the considerations set forth in the four factors, see Harper & Row, 471 U.S. at 549, 105 S.Ct. at 2225 (explaining that the four statutory fair use factors are nonexclusive), the District Court’s supplemental considerations are not actually supplemental, and as such should have instead been considered within the existing statutory framework.
The District Court’s first additional consideration was more properly considered under the first fair use factor — “the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes.” 17 U.S.C. § 107(1). Nonprofit educational uses are more likely to be fair because they promote the ultimate aims of copyright — the creation and dissemination of knowledge. Both of these aims must be kept in mind when evaluating a claim of fair use. Thus, whether the limited unpaid copying of excerpts will deter academic authors from creating is relevant. Nevertheless, it is the publishers — not academic authors — that are the holders of the copyrights at issue here. Publishers — not authors — are claiming infringement. Thus, when determining whether Defendants’ unpaid copying should be excused under the doctrine of fair use in this case, we are primarily concerned with the effect of Defendants’ copying on Plaintiffs’ incentive to publish, not on academic authors’ incentive to write.
The District Court’s second additional' consideration may be divided into two findings. First, the District Court found that a slight diminution of Plaintiffs’ permissions income caused by Defendants’ fair use would not appreciably harm Plaintiffs’ ability to publish scholarly works. However, this consideration is adequately dealt with under the fourth factor, which, as we have explained, asks whether the market harm caused by Defendants’ unpaid copying will materially impair Plaintiffs’ incentive to publish.
Second, the District Court found that that “it is consistent with the principles of copyright to apply the fair use doctrine in a way that promotes the dissemination of knowledge, and not simply its creation,” and treated this as a basis to allow the slight diminution of Plaintiffs’ permissions income caused by Defendants’ unpaid copying. Cambridge Univ. Press, 863 F.Supp.2d at 1241. We agree with the proposition that applying fair use in a manner which promotes the dissemination of knowledge is consistent with the goals of copyright. See Golan v. Holder, — U.S.-, 132 S.Ct. 873, 888, 181 L.Ed.2d 835 (2012) (explaining that the Progress of Science, which Congress is empowered to promote pursuant to the Copyright Clause, “refers broadly to the creation and spread of knowledge and learning” (emphasis added) (quotation marks omitted)). However, all unpaid copying could be said to promote the spread of knowledge, so this principal is not particularly helpful in “separating the fair use sheep from the infringing goats.” See Campbell, 510 U.S. at 586, 114 S.Ct. at 1175. To the extent that it is relevant here, this consideration is more neatly dealt with under the first factor, which teaches that educational uses— which certainly promote the dissemination of knowledge — are more likely to be fair. See § 107(1).
Accordingly, we hold that the District Court erred by separating the following considerations from its analysis of the first and fourth factors: (1) whether limited un*1283paid copying of excerpts will deter authors from creating new academic works, and (2) whether the slight limitation of permissions income caused by Defendants’ use would promote the spread of knowledge and would not appreciably dimmish Plaintiffs’ ability to publish scholarly works
E.
In sum, we hold that the District Court did not err in performing a work-by-work analysis of individual instances of alleged infringement in order to determine the need for injunctive relief. However, the District Court did err by giving each of the four fair use factors equal weight, and by treating the four factors mechanistically. The District Court should have undertaken a holistic analysis which carefully balanced the four factors in the manner we have explained.
The District Court did not err in holding that the first factor — the purpose and character of the use — favors fair use. Although Defendants’ use was nontransfor-mative, it was also for nonprofit educational purposes, which are favored under the fair use statute. However, the District Court did err in holding that the second fair use factor — the nature of the copyrighted work — favors fair use in every case. Though this factor is of comparatively little weight in this case, particularly because the works at issue are neither fictional nor ■ unpublished, where the excerpt in question contained evaluative, analytical, or subjectively descriptive material that surpasses the bare facts, or derive from the author’s own experiences or opinions, the District Court should have held that the second factor was neutral or even weighed against fair use where such material dominated.
With regard to the third factor — the amount used in relation to the copyrighted work as a whole — the District Court erred in setting a 10 percent-or-one-chapter benchmark. The District Court should have performed this analysis on a work-by-work basis, taking into account whether the amount taken — qualitatively and quantitatively — was reasonable in light of the pedagogical purpose of the use and the threat of market substitution. However, the District Court appropriately measured the amount copied based on the length of the entire book in all cases, declined to give much weight to the Classroom Guidelines, and found that the Defendants’ educational purpose may increase the amount of permissible copying.
With regard to the fourth factor — the effect of Defendants’ use on the market for the original — the District Court did not err. However, because Defendants’ unpaid copying was' nontransformative and they used Plaintiffs’ works for one of the purposes for which they are marketed, the threat of market substitution is severe. Therefore, the District Court should have afforded the fourth fair use factor more significant weight in its overall fair use analysis. Finally, the District Court erred by separating two considerations from its analysis of the first and fourth fair use factors, as described above supra part III. D.5.
Because the District Court’s grant of injunctive relief to Plaintiffs was predicated on its finding of infringement, which was in turn based on the District Court’s legally flawed methodology in balancing the four fair use factors and erroneous application of factors two and three, we find that the District Court abused its discretion in granting the injunction and the related declaratory relief. Similarly, because the District Court’s designation of Defendants as the prevailing party and consequent award of fees and costs were predicated on its erroneous fair use analysis, we find that the District Court erred in *1284designating Defendants as the prevailing party and awarding fees and costs to Defendants.
IV.
Accordingly, we REVERSE the judgment of the District Court. We VACATE the injunction, declaratory relief, and award of fees and costs to Defendants, and REMAND for further proceedings consistent with this opinion.
SO ORDERED.
. As the Association of Research Libraries recognizes, “[tjoday, students and teachers alike strongly prefer electronic equivalents (e-reserves for text, streaming for audio and video) to the old-media approaches to course , support.” Brief of American Library Association, et al., as Amicus Curiae Supporting Defendants/Appellees at 8 (quoting Association of Research Libraries, Code of Best Practices in Fair Use for Academic and Research Libraries 13 (2012), http://www.arl.org/storage/ documents/publications/code-of-best-practices-fair-use.pdf).
. Electronic reserve systems are popular not just at GSU, but nationwide. Many other university libraries offer electronic reserve systems to students and faculty that are similar to ERes and uLearn. See Brief of the Association of Southeastern Research Libraries as Amicus Curiae Supporting Defendants/Appellees at 3.
. In 2011, according to the Director of Digital Publishing Global for Cambridge, Frank Smith, "CCC [was] able to license excerpts from 60% of Cambridge's works.” Cambridge Univ. Press v. Becker, 863 F.Supp.2d 1190, 1213 (N.D.Ga.2012). Mr. Smith did not testify as to how many of Cambridge's works were available via CCC for licensed copying in 2009, the year during which the instances of copying that the District Court examined took place. In 2011, according to Oxford’s Acting President, Niko Pfund, "CCC ... license[d] the copying of excerpts from over 90% of Oxford's titles.” Id. Mr. Pfund did not testify as to how many of Oxford's works were available via CCC for licensed copying in 2009. However, the District Court found that "[t]here is documentary evidence showing that some but by no means all of these works were available for permissions in 2009.” Id. at 1214. In 2011, according to the Director of Licensing at Sage, Carol Richman, “CCC ... handle[d] permissions for all of Sage’s works.” Id. Ms. Richman did not testify as to how many of Sage's works were available via CCC for licensed copying in 2009. However, the District Court found that "documentary evidence shows that many though not all of Sage’s works at issue in this case generated permissions revenue from CCC during the period from July 1, 2004 to December 1, 2010.” Id.
. GSU stipulated that "[a]t GSU, coursepacks are printed and bound together and sold as units through the GSU bookstore.... GSU pays permission fees when copyrighted content is used in hardcopy coursepacks.” Doc. 276, at 8. Presumably this means that if no license for an excerpt of a particular work is available for use in a paper coursepack, GSU would simply not use that work. However, there is no evidence in the record that makes this explicit. Additionally, that GSU previously decided to pay for permissions to use paper excerpts does not necessarily weigh heavily on whether its current use of digital excerpts constitutes fair use. A number of factors might influence GSU’s decision whether to pay for permissions, or instead, claim fair use. As we make clear below in our analysis of the fourth fair use factor, the same logic also applies to Plaintiffs' decision whether to make digital permissions available. See infra note 32.
. In an action for direct infringement, the owner of a copyright may bring a claim against anyone who violates one of the “exclusive rights” protected by the Copyright Act. 17 U.S.C. § 501(a). The exclusive rights protected by the Copyright Act are as follows, subject to certain limitations set forth in other sections:
(1)to reproduce the copyrighted work in copies or phonorecords;
(2) to prepare derivative works based upon the copyrighted work;
(3) to distribute copies or phonorecords of the copyrighted work to the public by sale or other transfer of ownership, or by rental, lease, or lending;
(4) in the case of literary, musical, dramatic, and choreographic works, pantomimes, and motion pictures and other audiovisual works, to perform the copyrighted work publicly;
(5) in the case of literary, musical, dramatic, and choreographic works, pantomimes, and pictorial, graphic, or sculptural works, including the individual images of a motion picture or other audiovisual work, to display the copyrighted work publicly; and
(6) in the case of sound recordings, to perform the copyrighted work publicly by means of a digital audio transmission.
Id. § 106.
.A claim of contributory copyright infringement arises against one who intentionally induces or encourages the direct infringement of another. Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913, 930, 125 S.Ct. 2764, 2776, 162 L.Ed.2d 781 (2005). Although the Copyright Act does not expressly provide for any liability for infringement committed by another, "the[ ] doctrines of secondary liability emerged from common law principles and are well established in the law.” Id. (citations omitted).
. A claim of vicarious copyright infringement arises against one who "profit[s] from [another’s] direct infringement while declining to exercise a right to stop or limit it.” Id.
. GSU’s now-supplanted Regents’ Guide to Understanding Copyright & Educational Fair Use dates from 1997. According to a GSU professor who testified at trial, under the Regents' Guide GSU professors were routinely allowed to post digital copies of excerpts consisting of up to 20 percent of a work without obtaining a license from the copyright holder. Doc. 403, at 88:9-15.
. GSU’s 2009 Policy was modeled on the copyright policy of Columbia University, which employs a similar approach, including use of a four-factor fair use checklist. Appellees’ Br. at 9. The 2009 Policy is also similar to a policy jointly drafted in 2006 by Cornell University and the Association of American Publishers (the "AAP”) for the use of materials in Cornell’s electronic review system, after publishers represented by the AAP threatened to sue Cornell for copyright infringement. Brief of American Library Association, et al., as Amicus Curiae Supporting Defendants/Appellees at 3. Many other universities incorporate a fair use checklist as part of their copyright policy to assist instructors with making fair use determinations, including Duke University, Florida State University, University of Tennessee-Knoxville, and Louisiana State University. Brief of the Association of Southeastern Research Libraries as Amicus Curiae Supporting Defendants/Appellees at 9.
. For example, the Fair Use Checklist provides that the "[n]onprofit [e]ducational” use of an excerpt favors a finding of fair use, whereas “[c]ommercial activity” weighs against a finding of fair use. Doc. 235-2, at 7. Use of a ‘‘[fjactual or nonfiction work” favors a finding of fair use, whereas use of a "[h]ighly creative work (art, music, novels, films, plays, poetry, fiction)” weighs against a finding of fair use. Id. at 8. Use of a "[s]mall portion” of a work favors a finding of fair use, whereas use of a "[l]arge portion or entire work” weighs against a finding of fair use. Id. A use that has "[n]o significant effect on [the] market or potential market for [the] copyrighted work” favors a finding of fair use, whereas a use that '‘[s]ignficantly impairs [the] market or potential market for [the] copyrighted work or [a] derivative” weighs against a finding of fair use. Id.
. As the District Court’s Ex parte Young analysis in the following paragraph illustrates, however, despite the District Court’s preference for the word “indirect” — Plaintiffs’ direct infringement claim survived Defendants’ Motion to Dismiss.
. The Classroom Guidelines are the result of independent meetings that took place prior to passage of the Copyright Act of 1976 between "representatives of the Ad Hoc Committee of Educational Institutions and Organizations on Copyright Law Revision, and of the Authors League of America, Inc., and the Association *1246of American Publishers, Inc.," groups representing, respectively, educators, authors, and publishers. See H.R.Rep. No. 94-1476 at 67 (1976), reprinted in 1976 U.S.C.C.A.N. 5659, 5680-81. These meetings took place at the urging of the Chairman and other members of the House Judiciary Subcommittee, "in an effort to achieve a meeting of the minds as to permissible educational uses of copyrighted material.” Id., reprinted in 1976 U.S.C.C.A.N. 5659, 5680.
In a letter dated March 8, 1976, the three groups delivered to Congress an "Agreement on Guidelines for Classroom Copying in Not-For-Profit Educational Institutions with Respect to Books and Periodicals,” which came be known as the Classroom Guidelines. Id., reprinted in 1976 U.S.C.C.A.N. 5659, 5681. The Classroom Guidelines, which are reproduced in the legislative history of the 1976 Act, begin by stating that their “purpose ... is to state the minimum and not the maximum standards of educational fair use under Section 107.” Id. at 68, reprinted in 1976 U.S.C.C.A.N. 5659, 5681.
Under the terms of the Classroom Guidelines, teachers may make individual copies of excerpts of certain works for use in teaching, research, and class preparation, and may make multiple copies of works for classroom use so long as the copying meets the conditions of "brevity,” "spontaneity,” and "cumulative effect.” Id., reprinted in 1976 U.S.C.C.A.N. 5659, 5682. "Brevity” places strict word count limits on allowable copying, such as "an excerpt from any prose work of not more than 1,000 words or 10 percent of the work.” Id. "Spontaneity” provides that “[t]he inspiration and decision to use the work and the moment of its use for maximum teaching effectiveness are so close in time that it would be unreasonable to expect a timely reply to a request for permission.” Id. at 69, reprinted in 1976 U.S.C.C.A.N. 5659, 5682. “Cumulative effect” places caps on the total amount of copying permissible under the Classroom Guidelines. Id., reprinted in 1976 U.S.C.C.A.N. 5659, 5683. The Classroom Guidelines also contain other provisions, including prohibitions on charging students for copied materials beyond the cost of copying, copying that substitutes for the purchase of books and other publications, copying the same material in multiple terms, and copying that "substitute[s] for anthologies, compilations or collective works.” Id. at 69-70, reprinted in 1976 U.S.C.C.A.N. 5659, 5683.
"The [Classroom] [Guidelines were designed to give teachers direction as to the extent of permissible copying and to eliminate some of the doubt which had previously existed in this area of the copyright laws.... [T]hey are not controlling on the court.” Marcus v. Rowley, 695 F.2d 1171, 1178 (9th Cir.1983).
Whatever persuasive value the Classroom Guidelines may possess, we must keep in mind that they (1) were drafted by partisan groups, (2) "state the minimum and not the maximum standards of educational fair use under Section 107”, and (3) adopt the type of "hard evidentiary presumption[s]” regarding which types of use may be fair that the Supreme Court has since repeatedly warned against. See Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 584, 114 S.Ct. 1164, 1174, 127 L.Ed.2d 500 (1994). Indeed, the sentences immediately preceding the mechanistic "brevity,” "spontaneity,” and "cumulative effect” provisions illustrate the tentative nature of the groups’ recommendations:
The parties agree that the conditions determining the extent of permissible copying for educational purposes may change in the future; that certain types of copying permitted under these guidelines may not be permissible in the future; and conversely that in the future other types of copying not permitted under these guidelines may be permissible under revised guidelines. Moreover, the following statement of guidelines is not intended to limit the types of copying permitted under the standards of fair use under judicial decision and which are stated in Section 107 of the Copyright Revision Bill. There may be instances in which copying which does not fall within the guidelines stated below may nonetheless be permitted under the criteria of fair use.
H.R.Rep. No. 94-1476 at 67 (1976), reprinted in 1976 U.S.C.C.A.N. 5659, 5681.
. The District Court assumed that the Academic Annual Copyright License referred to CCC's Academic Repertory License Service, a subscription service which allows access to excerpts from a set group of works.
. Plaintiffs argue that the District Court's fair use analysis was legally deficient, but do not clarify whether they challenge the holding concerning only the 43 claims for which the District Court found no infringement, or whether they instead challenge all 48 claims for which the District Court undertook a fair use analysis. Because we review the legal sufficiency of the overall manner in which the District Court undertook its fair use analysis, we will not limit our review to the 43 claims for which the District Court found no infringement, but will consider all 48.
. Plaintiffs cite their first proposed injunction, filed May 11, 2011, modeled after the Classroom Guidelines. Appellants’ Br. at 83 (citing Doc. 300-1.).
. We note that " 'the Eleventh Amendment defense sufficiently partakes of the nature of a jurisdictional bar’ that it may be raised by the State for the first time on appeal,” Patsy v. Bd. of Regents of State of Fla., 457 U.S. 496, 515 n. 19, 102 S.Ct. 2557, 2567 n. 19, 73 L.Ed.2d 172 (1982) (quoting Edelman v. Jordan, 415 U.S. 651, 678, 94 S.Ct. 1347, 1363, 39 L.Ed.2d 662 (1974)). However, Eleventh Amendment immunity is not jurisdictional in the sense that we must consider it sua sponte, or must consider an argument that it applies that was not properly presented. See id. (observing that “the State may, under certain circumstances, waive th[e] [Eleventh Amendment] defense”).
. While it may seem counterintuitive to students of contemporary copyright, in eigh*1256teenth-century usage “science” meant "the field of human knowledge,” and "the useful arts" meant "technology and kindred know-how.” Thus, "science” refers to copyright while "useful arts” refers to patent. David L. Lange & Jefferson Powell, No Law: Intellectual Property in the Image of an Absolute First Amendment 36 (2009).
. The Supreme Court tells us that the idea/expression dichotomy and the fair use doctrine, coupled together, present the most important limitation on copyright law. See Eldred v. Ashcroft, 537 U.S. 186, 219-20, 123 S.Ct. 769, 788-89, 154 L.Ed.2d 683 (2003) ("[C]opyright law contains built-in First Amendment accommodations. First, it distinguishes between ideas and expression and makes only the latter eligible for copyright protection.... As we said in Harper & Row, this idea/expression dichotomy strikes a definitional balance between the First Amendment and the Copyright Act by permitting free communication of facts while still protecting an author’s expression. Due to this distinction, every idea, theoiy, and fact in a copyrighted work becomes instantly available for public exploitation at the moment of publication. Second, the fair use defense allows the public to use not only facts and ideas contained in a copyrighted work, but also expression itself in certain circumstances.” (citations omitted) (quotation marks omitted)). The Court reaffirmed this position most recently in Golan v. Holder, — U.S.-, 132 S.Ct. 873, 890-91, 181 L.Ed.2d 835 (2012) ("Given the speech-protective purposes and safeguards embraced by copyright law, we *1257concluded in Eldred that there was no call for the heightened [First Amendment] review petitioners sought in that case. We reach the same conclusion here [because the challenged law] leaves undisturbed the idea/expression distinction and the fair use defense.” (emphasis added) (footnotes omitted) (citations omitted) (quotation marks omitted)).
. Fair use was originally a common law doctrine that Congress codified in the Copyright Act of 1976. See 17 U.S.C. § 107.
. We understand "case-by-case” and "work-by-work” to be synonymous in cases where a copyright proprietor alleges numerous instances of copyright infringement and a sec*1259ondary user claims that his or her use was fair. Courts must apply the fair use factors to each work at issue. Otherwise, courts would have no principled method of determining whether a nebulous cloud of alleged infringements purportedly caused by a secondary user should be excused by the defense of fair use. See supra pages 1259-60.
The Second Circuit's approach in Cariou v. Prince, 714 F.3d 694 (2d Cir.2013) illustrates this point nicely. Cariou involved a relatively unknown photographer, Patrick Cariou; Richard Prince, a world famous appropriation artist; and the Gogosian Gallery in New York. Cariou spent six years taking photographic portraits of Rastafarians and landscape portraits of Jamaica. Cariou v. Prince, 784 F.Supp.2d 337, 344 (S.D.N.Y.2011). He then compiled these images into a book entitled Yes, Rasta. Id. Prince made a collage entitled Canal Zone out of 35 images from Cariou's Yes, Rasta collection, and ultimately completed a series of 30 Canal Zone paintings based on that collage. Id. at 343-45. Cariou sued Prince for copyright infringement. Id. at 353. Prince claimed each of his works constituted fair use. Id. The District Court evaluated each work individually and granted Cariou’s motion for summary judgment. Id. at 355. The Second Circuit, however, held that 25 of Prince’s Canal Zone paintings made fair use of Cariou's photographs and remanded the remaining five alleged instances of infringement to the District Court to consider each work more closely. Cariou, 714 F.3d at 708.
. Amici point out that excerpts in an electronic reserve system — like the ones at GSU — could be used for a different function than the original. For example, a scholarly monograph may be meant for use in research rather than teaching, and so use of that monograph in the classroom is arguably trans-formative. See Brief of American Library Association, et al., as Amici Curiae Supporting Defendants/Appellees at 14. Or, a professor “might assign an excerpt on the history of the civil rights movement by an author from Alabama and a similar excerpt by an author from New York not for the original purpose of conveying 'facts’ about these events but rather for the different purpose of discussing possible writer bias in the differing accounts of the events.” See Brief of the Association of Southeastern Research Libraries as Ami-cus Curiae Supporting Defendants/Appellees at 20.
We need not rule on whether such uses could ever be transformative, because the question is not before us. The District Court found that the works at issue in this case are targeted for use, at least in part, in the university classroom. Cambridge Univ. Press, 863 F.Supp.2d at 1211 n. 15. Because this finding is not clearly erroneous, we will not disturb it. Thus, Defendants’ use of excerpts of Plaintiffs’ works in university courses is for the same purpose — or one of the purposes, insofar as some of Plaintiffs’ works may also be meant for other purposes — as Plaintiffs' original works, and we need not consider the alternative scenarios amici suggest. Furthermore, we note that even if an individual work may be more in the vein of a research piece, the overall purpose of the works in question is nevertheless educational.
. This concern regarding the circularity of the analysis under the first factor as to "whether the user stands to profit from exploitation of the copyrighted material without paying the customary price” originates in an article written by the noted fair use scholar Judge Pierre Leval. See Pierre N. Leval, Nimmer Lecture: Fair Use Rescued, 44 UCLA L.Rev. 1449, 1460 (1997) (quoting Harper & Row Publishers, Inc. v. Nation Enters., 471 *1266U.S. 539, 562, 105 S.Ct. 2218, 2231, 85 L.Ed.2d 588 (1985)).
. Electronic reserve systems are not unique to GSU. As noted, "countless university libraries across the country” have such systems. Brief of the Association of Southeastern Research Libraries as Amicus Curiae Supporting Defendants/Appellees at 2. As such, the presence of such a system at GSU would hardly serve as a special draw to students or enhance GSU's reputation such as it might were it a unique advantage offered only at GSU.
. Although we agree with the thrust of the Second Circuit's characterization of the commercial/nonprofit dichotomy, we note that its assertion that "courts will not sustain a claimed defense of fair use” for a secondary use that amounts to "commercial exploitation,” Am. Geophysical Union v. Texaco Inc., 60 F.3d 913, 922 (2d Cir.1994) (emphasis added), would be better stated without resort to categorical language; i.e., courts will be less likely to sustain the fair use defense for secondary uses that represent commercial exploitation. To categorically exclude "commercial exploitation” from the purview of the fair use defense is contrary to the Supreme Court's direction that courts not adopt “hard evidentiary presumption[s]” with regard which types of use may be fair use. See Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 584, 114 S.Ct. 1164, 1174, 127 L.Ed.2d 500 (1994).
. As noted, copyright protects expression, and not the underlying ideas a work contains. See, e.g., Suntrust Bank v. Houghton Mifflin Co., 268 F.3d 1257, 1263 (11th Cir.2001). However, "the mere fact that the copied portions are themselves copyrightable cannot incline [the second] factor against fair use." 4 Melville B. Nimmer & David Nimmer, Nimmer on Copyright § 13.05[A][2][a] (2014). “[A] work will always be found 'original’ for copyrightability purposes before the fair use analysis is applied. The second statutory fair use factor ... refers to the 'nature' of the work beyond this initial inquiry.” Compaq Computer Corp. v. Ergonome Inc., 387 F.3d 403, 410 (5th Cir.2004) (holding that the second factor favored fair use because the work in question was factual despite the district court’s finding "at summary judgment that portions of [the work] were sufficiently original to warrant a copyrightability finding”), cert. denied, 544 U.S. 962, 125 S.Ct. 1742, 161 L.Ed.2d 604 (2005).
. As the District Court noted, the court in Princeton University Press appears to have been influenced by the defendants’ concession that the coursepacks contained creative material, and the parties there may have "confused an element of the prima facie case”— the threshold level of originality — with the *1269second factor analysis. Cambridge Univ. Press, 863 F.Supp.2d at 1227.
. For example, the District Court found that one of Plaintiffs’ works in question, The Organ as a Mirror of its Time: North European Reflections 1610-2000, "is a non-fiction work that discuss[es] various organs in Northern Europe over the past four centuries. The focus is on what organs reveal about the time in which they were built.” Cambridge Univ. Press, 863 F.Supp.2d at 1301. A GSU professor caused pages 78-91 of the book, representing one chapter entitled "The Organ in Seventeenth-Century Cosmology,” to be placed on ERes. Id. at 1300. The District Court held that "[t]he chapters are factual in nature.” Id. at 1301. However, if the chapter in question reflects the focus of the book, "what organs reveal about the time in which they were built,” id., then this type of creative analysis certainly goes beyond a mere compilation of facts regarding organs and represents the author’s individualized, subjective expression.
. The second factor is of comparatively little weight in this case, particularly because the works at issue are neither fictional nor unpublished. As to unpublished works, the Supreme Court discussed the circumstances under which authors have the right to decide when to reveal their work to the public in Harper & Row. As stated above, none of the works at issue in this case were unpublished; therefore we need not consider that matter further.
As to the fictional nature of a work however, the ground between the "paradigmatic example of a creative work” and the “bare factual compilation” is treacherous. This is so because courts who tread upon it must parse the pernicious idea/expression dichotomy to determine just how expressive a work is. Though, as mentioned supra note 25, the second statutory fair use factor refers to the nature of the work beyond the initial copy-rightability inquiry, copyrightability cases illustrate the difficulty one encounters when attempting to separate idea from expression.
For example, Judge Learned Hand once wrote in a copyright case involving a movie that was allegedly based on a play: “Upon any work ... a great number of patterns of increasing generality will fit equally well, as more and more of the incident is left out.... *1271rrjhere is a point in this series of abstractions where they are no longer protected, since otherwise the playwright could prevent the use of his ideas, to which, apart from their expression, his property is never extended. Nobody has ever been able to fix that boundary, and nobody ever can.” Nichols v. Universal Pictures, Corp., 45 F.2d 119, 121 (2d Cir.1930) (emphasis added); see also Bleistein v. Donaldson Lithographic Co., 188 U.S. 239, 251, 23 S.Ct. 298, 47 L.Ed. 460 (1903) (Holmes, J.) ("It would be a dangerous undertaking for persons trained only to the law to constitute themselves final judges of the worth of pictorial illustrations, outside the narrowest and most obvious of limits.”); Mannion v. Coors Brewing Co., 377 F.Supp.2d. 444, 459 (S.D.N.Y.2005) (noting "little is gained from attempting to distinguish an unprotectable idea from its protectable expression in a photograph or [any] other work of visual art”); Rebecca Tushnet, Worth a Thousand Words: The Image of Copyright, 125 Harv. L.Rev. 683, 702 (2012) ("The very excessive, worth-a-thousand-words quality of pictures may make them too unstable for courts accustomed to looking for meaning in words. With texts, by contrast, courts often feel more in control: courts have many standardized tools to interpret text, not least of all the rules of statutory and contractual construction. Those rules might be, in fact, indeterminate and manipulable, but they feel predictable and rational. For example, Judge Learned Hand's classic explanation of copyright’s idea/expression dichotomy [in Nichols] acknowledged that the distinction between the two is inherently arbitrary, but he was nonetheless perfectly comfortable applying it to written texts such as plays and screenplays.”).
. For example, the Classroom Guidelines’ “brevity” requirement allows copying of "an excerpt from any prose work of not more than 1,000 words or 10 percent of the work.” See supra note 12. Plaintiffs point out that the District Court consistently found Defendants' copying far in excess of 1,000 words to be fair use.
. There are ample precedents that explain that excessive verbatim copying weighs against fair use under factor three. See, e.g., Marcus v. Rowley, 695 F.2d 1171, 1176 (9th Cir.1983) ("[WJholesale copying of copyrighted material precludes application of the fair use doctrine.”); Walt Disney Prods, v. Air Pirates, 581 F.2d 751 (9th Cir.1978) ("Excessive copying precludes fair use....”); Rosemont Enters., Inc. v. Random House, Inc., 366 F.2d 303, 310 (2d Cir.1966) ("The fair use privilege is based on the concept of reasonableness and extensive verbatim copying or paraphrasing of material set down by another cannot satisfy that standard.”); Wihtol v. Crow, 309 F.2d 777, 780 (8th Cir.1962) (“[I]t is not conceivable to us that the copying of all, or substantially all, of a copyrighted song can be a ‘fair *1275use' merely because the infringer had no intent to infringe.”). Judge Chin, however, in Authors Guild, Inc. v. Google, Inc., 954 F.Supp.2d 282, 292 (S.D.N.Y.2013) explained the broad contours of the third fair use factor well when he wrote: "[On the one hand] Google scans the full text of books — the entire books — and it copies verbatim expression. On the other hand, courts have held that copying the entirety of a work may still be fair use. See, e.g., Sony Corp. of Am. v. Universal City Studios, Inc., 464 U.S. 417, 449-50, 104 S.Ct. 774, 78 L.Ed.2d 574 (1984); Bill Graham Archives, 448 F.3d at 613 ("[C]opying the entirety of a work is sometimes necessary to make a fair use of the image[.]”).”
. At one point, the Supreme Court described the fourth factor as "the single most important element of fair use,” Harper & Row, 471 U.S. at 566, 105 S.Ct. at 2233. However, in Campbell, the Supreme Court stated that the fair use analysis "is not to be simplified with brightline rules, for the statute, like the doctrine it recognizes, calls for case-by-case analysis .... Nor may the four statutory factors be treated in isolation, one from another. All are to be explored, and the results weighed together, in light of the purposes of copyright.” 510 U.S. 569, 577-78, 114 S.Ct. 1164, 1170-71, 127 L.Ed.2d 500 (1994). This language appears to be inconsistent with any single factor being deemed the single most important. In its discussion of the fourth factor in Campbell, the Supreme Court conspicuously did not single it out as having any greater significance than the others. 510 U.S. at 590-94, 114 S.Ct. at 1177-79. Many courts and commentators, following Campbell, have determined that the fourth factor *1276should not be given primacy. E.g., Texaco, 60 F.3d at 926; William F. Patry, Patry on Fair Use 450-56 (2011).
Nevertheless, because Defendants’ use of Plaintiffs’ works is nontransformative and hence the threat of market substitution is severe, it is appropriate in this instance to afford relatively great weight to the fourth factor in the overall fair use analysis.
. Of course, it need not always be true that a publisher’s decision not to make a work available for digital permissions conclusively establishes that the publisher envisioned little or no demand, and that the value of the permissions market is zero. After all, a number of other factors might influence a publisher’s distribution decision: the publisher may not yet have figured out how to sell work in a different medium, or it might want to restrict circulation in one medium to promote another.
. Plaintiffs argue that the District Court improperly focused on the availability of digital licenses in 2009. Such availability, Plaintiffs point out, is irrelevant because Plaintiffs seek only prospective injunctive relief, and the status of licensing in 2009 has no bearing on the question of whether GSU’s copyright policy going forward should require an investigation of whether a license is available.
However, the fair use analysis must be performed on a work-by-work basis. Campbell, 510 U.S. at 577, 114 S.Ct. at 1170. As such, the District Court had to examine specific instances of infringement in order to determine whether prospective relief was warranted, and what shape that relief should take. In deciding those individual cases, the availability of licensing at the time of an alleged infringement — not at some undefined time in the future — is the relevant evidence.
Plaintiffs also complain that they were not notified at trial that they were required to show license availability for 2009 in particular. However, Plaintiffs were aware that the trial was to be conducted on the basis of the individual instances of purported infringement taking place in 2009 that were enumerated in the March 15, 2011, joint document. Thus, Plaintiffs were — or should have been— aware that the relevant evidentiary question was whether a market existed at the time of the purported infringements.
. We note that placing the burden of going forward with the evidence regarding license availability on Plaintiffs does not create a proscribed "hard evidentiary presumption” regarding categories of use that are fair use against. See Campbell, 510 U.S. at 584, 114 S.Ct. at 1174. Placing this burden on Plaintiffs creates no presumption about whether a given instance of copying will be fair use. This approach merely recognizes that this is a case wherein one party has all the evidence on a particular issue, and so it is equitable to require that party to go forward with the evidence.
. Plaintiffs did not put on evidence demonstrating that large numbers of universities other than GSU were making unpaid use in electronic reserves of the works that are the subject of this case. Nevertheless, the District Court did not need such evidence in order to properly find that widespread conduct of the type engaged in by GSU would cause substantial harm to the market for Plaintiffs’ works because the Copyright Act directs us to consider the effect on the potential market. See § 107(4).
We note that Plaintiffs could have framed their case differently by putting on evidence of copying at other universities in order to show the negative impact a finding of fair use would have on the existing market for the works in question. Because this is not the case before us, however, we need not speculate as to its outcome
. Defendants argue that the District Court should have gone further and shifted the ultimate burden of persuasion on market harm to Plaintiffs. Defendants argue that support for this position may be found in Sony, in which the Supreme Court placed the burden on the holders of copyrights on television programs to demonstrate that noncommercial recording by home users of Sony's videotape recorders likely caused nonminimal harm to the market for their works. 464 U.S. at 451, 104 S.Ct. at 793. Defendants contend that Sony created a rule that the overall burden of proof on the fourth factor is reversed in cases of noncommercial use, and argue that other courts have interpreted Sony in this manner. See Princeton Univ. Press, 99 F.3d at 1385-86 (“The burden of proof as to market effect rests with the copyright holder if the challenged use is of a 'noncommercial' nature. The alleged in-fringer has the burden, on the other hand, if the challenged use is ‘commercial’ in nature.”).
We are not convinced that Sony stands for such a proposition. The portion of Sony Defendants cite discusses a presumption that commercial uses are not fair use and noncommercial uses are fair use. See Sony, 464 U.S. at 449-51, 104 S.Ct. at 792-93. The Supreme Court has since clarified that such presumptions have no place in the fair use analysis. See Campbell, 510 U.S. at 584, 114 *1281S.Ct. at 1174. Campbell’s treatment of Sony suggests- that Sony’s discussion of presumptions and burdens is better viewed as a discourse on the common sense inferences that may be drawn regarding the threat of market substitution in cases of nontransformative use (although at the time of Sony the Court had not yet adopted that term).
Furthermore, in Campbell, the Supreme Court stated unequivocally that "[s]ince fair use is an affirmative defense, its proponent would have difficulty carrying the burden of demonstrating fair use without favorable evidence about relevant markets.” 510 U.S. at 590, 114 S.Ct. at 1177 (footnotes omitted). Defendants contend that the Campbell court was dealing, with commercial use and did not disapprove Sony’s shifting of the burden in noncommercial cases. However, we find that the Court’s unqualified statement that fair use is an affirmative defense in its discussion of the fourth factor indicates that the Court did not intend for the overall burden of proof to be reversed.
. The 17 works for which Plaintiffs failed to provide evidence of digital license availability were published by Oxford and Cambridge. Thus, evidence concerning Sage is not relevant to the question of whether the District Court erred in finding that mo licenses were available for those works.