OPINION AND ORDER
PHILIP P. SIMON, District Judge.Plaintiff Brenda D. Jones appeals the Social Security Administration’s decision to deny her application for Supplemental Security Income. An Administrative Law Judge found that Jones was not disabled within the meaning of the Social Security Act. As explained below, I find that the ALJ failed to give proper weight to the opinion of Jones’ treating psychiatrist with respect to the severity of her mental impairments, and I will REMAND the matter back to the ALJ for further proceedings consistent with this opinion
Background
Jones, who is now 47 years old, has no shortage of physical and mental impairments. She has aortic sclerosis, kidney stenosis, degenerative disc disease, and hypertension [DE 8 at 26-27]. She suffers from migraines and occipital neuralgia, and has struggled with dependency on marijuana and prescription medication [Id.]. But what’s important for our purposes is that Jones has been diagnosed as having a bipolar subtype of schizoaffective disorder [DE 8 at 27, 644], Schizoaffective disorder combines features of both schizophrenia and a mood disorder [DE 8 at 695]. In Jones’ case, this means auditory hallucinations (hearing voices) and periods of depression alternating with wild mood swings and explosions of anger [Id.].
Jones filed applications for Title II social security disability benefits and Title XVI supplemental security income benefits in May 2010, but these were initially denied by the SSA [DE 8 at 24]. Jones dropped her claim for Title II benefits, but persisted in the claim for SSI benefits. After a hearing, an ALJ upheld the agency’s denial of SSI benefits [Id.]. The denial is now before me on review.
Unlike most social security appeals, Jones makes only one argument: the ALJ *876erred by failing to credit the opinion of one of her treating psychiatrists, Dr. Michael Conn, M.D., when determining Jones’ residual functional capacity [DE 16 at 11-13].1 More concretely, the ALJ disregarded Dr. Conn’s opinion that Jones would miss more than three days of work a month and could only remain on-task for less than 85% of a workday [DE 8 at 696-97].
This is important because a vocational expert who testified at the hearing concluded that, try as they might, a person who missed three days of work a month would not be able to maintain a job [DE 8 at 80]. The same goes for someone who could only stay on-task for less than 90% of the workday [Id.]. In other words, if you believe Dr. Conn, then Jones’ application probably shouldn’t have been denied.
The ALJ didn’t believe Dr. Conn, however, and I have to determine whether he had good reasons not to. The ALJ’s first reason is that Dr. Conn’s statement was too general. It addressed people with schizoaffective disorder in the abstract rather than Jones specifically. In addition, the ALJ believed that Dr. Conn’s opinion was undercut by medical records showing a Global Assessment of Functioning (GAF) score of 58 because that score indicates only moderate difficulty in social or occupational functioning. As I’ll explain in more detail below, neither of these reasons are good enough to justify rejecting Dr. Conn’s assessment.
Discussion
If an ALJ’s findings of fact are supported by “substantial evidence” then they must be sustained. See 42 U.S.C. § 405(g). Substantial evidence consists of “such relevant evidence as a reasonable mind might accept as adequate to support a conclusion.” Nelms v. Astrue, 553 F.3d 1093, 1097 (7th Cir.2009) (quoting Richardson v. Perales, 402 U.S. 389, 399-400, 91 S.Ct. 1420, 28 L.Ed.2d 842 (1971)). Review of the ALJ’s findings is deferential. See Overman v. Astrue, 546 F.3d 456, 462 (7th Cir.2008). In making a substantial evidence determination, I must review the record as a whole, but I can’t re-weigh the evidence or substitute my judgment for that of the ALJ. Id.
“Although this standard is generous, it is not entirely uncritical.” Id. at 462 (quoting Steele v. Barnhart, 290 F.3d 936, 940 (7th Cir.2002)). I must ensure that the ALJ has built a “logical bridge” between the evidence and the result. See Getch v. Astrue, 539 F.3d 473, 481 (7th Cir.2008). However, if reasonable minds could differ on whether a claimant is disabled, I should affirm the decision denying benefits. See Elder v. Astrue, 529 F.3d 408, 413 (7th Cir.2008).
The key to this case is the fact that Dr. Conn was Jones’ treating psychiatrist. He saw her frequently, once every couple of months between March 2010 and January 2012 [DE 8 at 430-34; 551-554; 628-700]. This is important because the SSA regulations require that the opinion of a treating psychiatrist be given controlling weight if it is “well-supported by the medically acceptable clinical and laboratory diagnostic techniques and is not inconsistent with the other substantial evidence.” 20 C.F.R. § 404.1527(c)(2); accord Roddy v. Astrue, 705 F.3d 631, 636 (7th Cir.2013). If an ALJ decides to reject a treating source’s opinion, he is required to provide a sound explanation for that decision. Roddy, 705 F.3d at 636; Jelinek v. Astrue, 662 F.3d 805, 811 (7th Cir.2011); Punzio v. Astrue, 630 F.3d 704, 710 (7th Cir.2011).
*877The opinion at issue is a Medical Source Statement Dr. Conn completed in January 2012 [DE 8 at 695-700]. In it, Dr. Conn opined that Jones’ schizoaffective disorder would cause her to miss at least three workdays a month and would prevent her from remaining on-task for more than 85% of a workday. The ALJ gave Dr. Conn’s opinion only limited weight, and made two arguments justifying that decision. Neither is sound.
First, the ALJ argued that Dr. Conn’s opinion was inappropriately general because the doctor merely “described ‘people with schizoaffective disorder’ throughout his statement” rather than “talking about the claimant” [DE 8 at 35]. As the government acknowledges in their brief, this argument doesn’t stand up to scrutiny [DE 24 at 4]. Dr. Conn describes Jones’ specific symptoms and limitations at length, explaining that Jones reported hearing voices on a daily basis and that her mood swings ranged from spells of poor energy, crying, and irritability to manic stages consisting of violent outbursts [DE 8 at 695]. While Dr. Conn does occasionally write generally about people with schizoaffective disorder, he does so in the context of explaining how the disorder affects Jones. For instance, when explaining why she would have trouble maintaining attention during a workday, Dr. Conn writes generally about auditory hallucinations and mood disorders, but then specifically explains that Jones hears demeaning voices in her head that would distract her from her work [DE 8 at 697]. So the ALJ’s first reason is simply not a sound one.
Nor is the second. The second argument advanced by the ALJ is that Dr. Conn’s opinion is inconsistent with medical records showing a GAF score of 58 [DE 8 at 35]. This argument doesn’t hold up for two related reasons. First, a GAF score is a diagnostic tool-it represents the clinician’s judgment of the individual’s overall level of psychological, social, and occupational functioning. Denton v. Astrue, 596 F.3d 419, 425 (7th Cir.2010) (per curiam). Although superficially similar, it is not the equivalent of a doctor’s opinion of functional capacity and is not treated as such by the regulations. Id. (“nowhere do the Social Security regulations or case law require an ALJ to determine the extent of an individual’s disability based entirely on his GAF score”). Jones’ score of 58 means, on those occasions, she was either exhibiting moderate symptoms of mental illness or experiencing moderate difficulty in social or occupational functioning. See Punzio, 630 F.3d at 711. But moderate symptoms and moderate difficulties are not inconsistent with the inability to hold down a job. After all, “there can be a great distance between a patient who responds to treatment and one who is able to enter the workforce” Scott v. Astrue, 647 F.3d 734, 739 (7th Cir.2011). Dr. Conn apparently believed both that Jones was responding well to treatment and that she would be unable to hold a job. These are not necessarily contradictory beliefs. See Punzio, 630 F.3d at 711 (a GAF score of 60 was not at odds with a finding that a claimant was unable to hold a job).
Even if the GAF scores did accurately reflect Jones’ actual functional capacity, the scores still would not be a sound reason for rejecting Dr. Conn’s opinion. The nature of mental illness is that people experience fluctuations in symptoms. So records showing a patient had a “good day” or is feeling better do not imply that the underlying condition has disappeared. Scott, 647 F.3d at 740; Punzio, 630 F.3d at 710 (“a person who suffers from a mental illness will have better days and worse days, so a snapshot of any single moment says little about her overall condition.”). Here the record shows the type of mixed picture one would expect from someone suffering from a bipolar condition. Much of the time, Dr. Conn reported that Jones *878was generally euthymic (non-depressed), stable, and not anxious [DE 8 at 430, 554, 631, 635, 648]. However, at other times he reported Jones as having “poor energy and frequent crying spells,” feeling “irritable and weepy,” talking to herself and hearing voices frequently [DE 8 at 432, 644]. Jones hit a particularly low point in September 2011 when she was arrested for disorderly conduct after an angry confrontation with her family. At the next counseling session she appeared disheveled and anxious and reported daily auditory hallucinations [DE 8 at 628-29]. Focusing on two good GAF scores out of a mixed record represents a type of “cherry-picking” that the Seventh Circuit has repeatedly warned about. See Punzio, 630 F.3d at 710 (cherry picking the record of a mentally ill claimant for good days demonstrates a “regrettably all-too-eommon misunderstanding of mental illness”); Scott, 647 F.3d at 740 (vacating an ALJ’s decision denying benefits to bipolar claimant because the ALJ cherry-picked positive treatment notes instead of considering the entire record.)
Those were the two reasons the ALJ gave for discounting Dr. Conn’s opinion. The Commissioner advances a few more in her brief, arguing Dr. Conn’s opinion was biased, was based on Jones’ unreliable self-reported symptoms, and that certain medical records cast doubt on Dr. Conn’s assessment [DE 24 at 5-8]. These arguments run afoul of the Chenery doctrine which prohibits an agency’s lawyers from defending the agency’s decision on grounds that the agency itself had not embraced. See SEC v. Chenery Corp., 318 U.S. 80, 87, 63 S.Ct. 454, 87 L.Ed. 626 (1943) (“The grounds upon which an administrative order must be judged are those upon which the record discloses that its action was based.”); see also Kastner v. Astrue, 697 F.3d 642, 648 (7th Cir.2012) (“Under the Chenery doctrine, the Commissioner’s lawyers cannot defend the agency’s decision on grounds that the agency itself did not embrace. On appeal, the Commissioner may not generate a novel basis for the ALJ’s determination.”); Mendez v. Barnhart, 439 F.3d 360, 362 (7th Cir.2006).
The ALJ made some of these points when discussing other evidence in the record, but did not rely on them to discount Dr. Conn’s January 2012 opinion. Therefore the Commissioner can’t use these arguments to justify the ALJ’s decision. The Commissioner attempts to evade the strictures of Chenery by claiming that these arguments can be inferred from the ALJ’s decision [DE 24 at 5]. But even if they are somehow implicit in the decision, the ALJ had a duty to make them explicit — to connect the evidence to his conclusion through an “accurate and logical bridge.” Stewart v. Astrue, 561 F.3d 679, 684 (7th Cir.2009); Berger v. Astrue, 516 F.3d 539, 544 (7th Cir.2008). This he did not do, so I am not going to address the merits of these arguments here. On remand, the ALJ will have the opportunity to revisit the Commissioner’s arguments against Dr. Conn’s opinion and adopt them if he thinks they have merit.
To sum up: the ALJ needed solid reasons for rejecting Dr. Conn’s opinion and the ALJ either didn’t have them or didn’t articulate them. This failure is fatal and must be reversed. Scott, 647 F.3d at 740-41 (reversing decision for failure to credit a treating psychiatrist’s assessment of a claimant’s mental illness).
CONCLUSION
For the reasons stated above, this cause is REMANDED for further proceedings consistent with this order.
SO ORDERED.
. Assessing an applicant's residual functional capacity is one of the five "steps” than an ALJ must undertake to determine eligibility. See Winfield v. Comm’r of Soc. Sec., No. 2:11—cv432-PPS, 2013 WL 692408, at *3 (N.D.Ind. Feb. 25, 2013) (explaining the "step” terminology).