Opinion by Judge IKUTA; Partial Concurrence and Partial Dissent by Judge GRABER.
OPINION
IKUTA, Circuit Judge:Edilia Molina appeals the district court’s decision affirming the Social Security Commissioner’s denial of her application for disability insurance benefits and supplemental security income under the Social Security Act. She argues that the administrative law judge (ALJ) erred by giving *1108inadequate weight to the opinions of her primary care provider, by improperly determining she was not credible, and by rejecting the testimony of her family members without comment. We have jurisdiction pursuant to 28 U.S.C. § 1291, and we affirm.
I
On September 22, 2006, Molina filed claims for disability benefits and supplemental security income, alleging that she suffered from panic and anxiety attacks1 and had been unable to work since September 7, 2002, when she last held a job as an assembler for Texas Instruments. According to Molina’s primary care provider, physician’s assistant Molly Wheelwright, Molina had a “longstanding panic disorder,” and had tried several anxiety medications and antidepressants, including Lorazepam, over the years. Wheelwright also noted that Molina had to leave work and visit the emergency room more than once because of “severe panic attacks.” After September 7, 2002, Molina’s alleged disability onset date, Wheelwright’s records showed that Molina continued to have symptoms of panic disorder, although as of February 2005, Molina stated she was stable on Lorazepam. Although Wheelwright frequently encouraged Molina to seek counseling for her anxiety, Molina declined to do so, citing various concerns including unpleasant past experiences with psychiatrists and a belief that the local mental health clinic was for drug addicts.
A few weeks after Molina filed her application with the Social Security Administration, she contacted a clinician at a nearby mental health clinic for intake and evaluation. The clinician wrote that Molina was “motivated” as well as “calm, orderly, [and] normally responsive.” Molina was diagnosed with post-traumatic stress disorder and panic disorder. She failed to show up for her follow-up appointment, and told a clinician in June 2007 that she no longer wanted counseling services.
Dr. Hunter Yost, a psychiatrist, examined Molina in November 2006 and reported that Molina “ha[d] some partial symptoms” of anxiety. After conducting a mental status examination, Dr. Yost concluded that Molina was alert and oriented, maintained good eye contact, and “did not appear excessively anxious.” She spoke coherently and fluently, smiled appropriately, and was congenial. She did not have an elated or depressed mood, and there was no evidence of a thought disorder. Dr. Yost diagnosed her with panic disorder without agoraphobia, and wrote that her condition was controlled with Lorazepam and that she was able to do self-calming measures to help with panic attacks when she was out in public. He concluded that her disorder had not and would not impose any limitations on her ability to work. A state Disability Determination Services staff physician reviewed Molina’s file in December 2006 and reached similar conclusions.
In December 2006, Wheelwright wrote that Molina was feeling generally well and described her anxiety episodes as intermittent. In February 2008, however, in a Mental Impairment Questionnaire for Molina’s disability application, Wheelwright wrote that Molina had “panic disorder, agoraphobia” with an “unsure” prognosis. In the grid for “mental abilities and aptitudes needed to do unskilled work,” Wheelwright checked “unable to meet competitive standards” next to “maintain regular attendance,” “deal with normal work stress,” and “complete a normal workday and workweek without interruptions from psychologically based symp*1109toms,” among other categories. Although the form instructed her to explain these assessments and provide medical or clinical findings to support them, Wheelwright did not do so. As for Molina’s functional limitations, Wheelwright checked “extreme” next to the boxes for “restriction of activities of daily living,” “difficulties in maintaining social functioning,” and “difficulties in maintaining concentration, persistence, or pace.” In separate medical notes, Wheelwright wrote that Molina was “very disabled by her condition,” because she “[b]ecomes easily panicked in many situations even throughout the day” and is “[ujnable to work as a result.” She noted, however, that a psychiatrist would “likely need to supplement” her evaluation.
At her hearing before the ALJ, Molina testified that she had been helping care for her two grandchildren since 2003, including walking them to school in the morning and picking them up in the afternoon. Molina had cared for the younger granddaughter full-time from 2003 until she was old enough to start school in 2005. A state agency paid her for her services until 2007. Molina said she regularly attended church, went shopping with her sister, attended family barbeques, and went for short walks in the neighborhood. Molina also submitted two function reports in which she wrote that she drove a car in the immediate area and went shopping once a month but liked to get in and out quickly because she did not like to be around a lot of people. She drove when she had to but felt a lot of anxiety and would “only go very short distances.” She went out to eat with friends or family occasionally, but once her condition began, she would experience panic attacks that made it difficult for her to swallow her food and would have to go outside and wait for the others to finish their meals. She did not like to go out because she did not like to be around a lot of people and would experience panic attacks; these panic attacks were characterized by anxiety, shortness of breath, dizziness, sweating, and an elevated heart rate. She liked to watch TV and eat dinner in her room.
Molina testified that her last job as an assembler for Texas Instruments involved performing wire bonding and inspection. She worked in a large room with four or five other people, and her job did not require her to speak to anyone unless there was a problem with her machine, in which case she would ask the setup operator for help. At her lunch break she would take her food to her car and eat by herself. She said she was not able to return to this job because she could not tolerate the presence of other people. When she worked, she would get anxiety and panic attacks and would have to go to the infirmary to lie down or to the hospital. These attacks would come on without warning; her heart would start to beat very hard, and she would have difficulty breathing and feel fearful.
Five family members submitted statements on Molina’s behalf. Molina’s sister Mary stated that Molina was scared to get out on her own, could not stand to be around a lot of people, and did not feel safe by herself. She would go to her room when people came to the house. Her sister Frances De La Cerda wrote that Molina’s anxiety was “so severe that she is not able to go anywhere without having a panic attack” and that she “had to quit working because the anxiety affected her mental abilities.” According to her brother-in-law, David Chenoweth, Molina did “not like to travel anywhere except for short trips in the daytime,” and she seemed to be most comfortable in her room. Her sister Barbara Molina wrote that Molina could not be around people or go out to eat because she tended to get very nervous and would take her food home with her. Molina’s sister-in-law, Theresa Molina, *1110said that when Molina had a panic attack, she would get very shaky and sweaty and feel faint, and that she preferred not to go anywhere for fear that she would have one of her attacks.
Following the hearing, the ALJ issued a written opinion holding that Molina was not disabled because she could perform her past relevant work as an assembler. In reaching this conclusion, the ALJ discussed Dr. Yost’s evaluation, the Mental Impairment Questionnaire filled out by Wheelwright, and Molina’s testimony at the hearing. The ALJ adopted Dr. Yost’s conclusion regarding the severity of Molina’s mental impairments. She found that Molina’s subjective allegations regarding the severity of her pain and ability to work were not credible to the extent they conflicted with Dr. Yost’s evaluation. She considered but discounted Wdieelwright’s opinions because they were “quite conclusory” and provided “very little explanation of the evidence relied on in forming those opinions.” Additionally, they did not appear to be supported by Molina’s objective medical condition and were inconsistent with Dr. Yost’s opinion. The ALJ also noted that Molina had cared for her granddaughters throughout much of the alleged disability period, and that she had not made any efforts to seek counseling from 2003 to 2005. The district court affirmed.
On appeal, Molina argues that the ALJ erred in reaching this conclusion because (1) the ALJ should have given more weight to Wheelwright's opinions, (2) the ALJ improperly evaluated Molina’s credibility, and (3) the ALJ erred in rejecting the testimony of Molina’s family members without comment. We consider each of these arguments in turn.
II
For purposes of the Social Security Act, a claimant is disabled if the claimant is unable “to engage in any substantial gainful activity by reason of any medically determinable physical or mental impairment which can be expected to result in death or which has lasted or can be expected to last for a continuous period of not less than 12 months.” 42 U.S.C. § 423(d)(1)(A). In order to determine whether a claimant meets this definition, the ALJ employs a five-step sequential evaluation. Parra v. Astrue, 481 F.3d 742, 746 (9th Cir.2007); 20 C.F.R. §§ 404.1520(a), 416.920(a). In brief, the ALJ considers whether a claimant is disabled by determining: (1) whether the claimant is “doing substantial gainful activity”; (2) whether the claimant has a “severe medically determinable physical or mental impairment” or combination of impairments that has lasted for more than 12 months; (3) whether the impairment “meets or equals” one of the listings in the regulations; (4) whether, given the claimant’s “residual functional capacity,” the claimant can still do his or her “past relevant work”; and (5) whether the claimant “can make an adjustment to other work.” 20 C.F.R. §§ 404.1520(a), 416.920(a). The claimant bears the burden of proof at steps one through four. Parra, 481 F.3d at 746.
We review the district court’s order affirming the ALJ’s denial of social security benefits de novo, Tommasetti v. Astrue, 533 F.3d 1035, 1038 (9th Cir.2008), and reverse only if the ALJ’s decision was not supported by substantial evidence in the record as a whole or if the ALJ applied the wrong legal standard, Stone v. Heckler, 761 F.2d 530, 531 (9th Cir.1985). Substantial evidence “ ‘means such relevant evidence as a reasonable mind might accept as adequate to support a conclusion.’ ” Valentine v. Comm’r Soc. Sec. Admin., 574 F.3d 685, 690 (9th Cir.2009) (quoting Desrosiers v. Sec’y of Health & Human Servs., 846 F.2d 573, 576 (9th Cir.1988)). The evidence must be “more than a mere *1111scintilla,” but may be less than a preponderance. Id. (quoting Desrosiers, 846 F.2d at 576). Even when the evidence is susceptible to more than one rational interpretation, we must uphold the ALJ’s findings if they are supported by inferences reasonably drawn from the record. Tommasetti, 533 F.3d at 1038. Finally, we may not reverse an ALJ’s decision on account of an error that is harmless. See Stout v. Comm’r, Soc. Sec. Admin., 454 F.3d 1050, 1055-56 (9th Cir.2006). “[T]he burden of showing that an error is harmful normally falls upon the party attacking the agency’s determination.” Shinseki v. Sanders, 556 U.S. 396, 409, 129 S.Ct. 1696, 173 L.Ed.2d 532 (2009).
Ill
We first turn to Molina’s argument that the ALJ erred in discounting Wheelwright’s opinions about Molina’s residual functional capacity. The ALJ rejected Wheelwright’s views on the grounds that Wheelwright’s opinions were “quite conclusory,” provided very little explanation of the evidence relied on, were not supported by Molina’s objective medical condition, and were inconsistent with the opinion of Dr. Yost, the examining psychiatrist.
In order to reject the testimony of a medically acceptable treating source, the ALJ must provide specific, legitimate reasons based on substantial evidence in the record. Valentine, 574 F.3d at 692. However, only licensed physicians and certain other qualified specialists2 are considered “ [acceptable medical sources.” 20 C.F.R. § 404.1513(a). Physician’s assistants are defined as “other sources,” § 404.1513(d), and are not entitled to the same deference, see § 404.1527; SSR 06-03p. The ALJ may discount testimony from these “other sources” if the ALJ “ ‘gives reasons germane to each witness for doing so.’ ” See Turner v. Comm’r of Soc. Sec., 613 F.3d 1217, 1224 (9th Cir.2010) (quoting Lewis v. Apfel, 236 F.3d 503, 511 (9th Cir.2001)).
Wheelwright did not qualify as a medically acceptable treating source because she was a physician’s assistant, see 20 C.F.R. § 404.1513(d)(1), and the record does not show that she worked under a physician’s close supervision, see Gomez v. Chater, 74 F.3d 967, 971 (9th Cir.1996) (holding that a nurse practitioner could be considered a medically acceptable source where she worked under a physician’s close supervision such that she acted as the physician’s agent).3 The ALJ gave several germane reasons for discounting Wheelwright’s opinions in favor of the conflicting testimony of Dr. Yost, and these reasons were substantiated by the record. First, Wheelwright’s opinion as to the impact of Molina’s mental impairments on her ability to work consisted primarily of a standardized, check-the-box form in which she failed to provide supporting reasoning or clinical findings, despite being instructed to do so. We have held that the ALJ may “permissibly reject[ ] ... check-off reports that [do] not contain any explanation of the bases of their conclusions.” Crane v. Shalala, 76 F.3d 251, 253 (9th Cir.1996); see also Holohan v. Massanari, 246 F.3d 1195, 1202 (9th Cir.2001) (“[T]he *1112regulations give more weight to opinions that are explained than to those that are not.”). Second, Wheelwright’s opinions that Molina “[bjecomes easily panicked in many situations even throughout the day” and is “unable to work as a result,” were conclusory and conflicted with her earlier assessment that Molina’s anxiety attacks were only intermittent. Finally, as the ALJ held, Wheelwright’s opinion was inconsistent with that of Dr. Yost, who specialized in the relevant field of psychiatry, and whose opinion was therefore entitled to greater weight. See Holohan, 246 F.3d at 1202 (“[T]he regulations give more weight to ... the opinions of specialists concerning matters relating to their specialty over that of nonspecialists.”); Smolen v. Chater, 80 F.3d 1273, 1285 (9th Cir.1996) (holding that the ALJ should have given greater weight to a physician with the expertise that was most relevant to the patient’s allegedly disabling condition). In fact, Wheelwright herself indicated that her evaluation would need to be supplemented by a psychiatrist. Accordingly, the ALJ did not err in discounting Wheelwright’s opinion where it conflicted with Dr. Yost’s evaluation.4
IV
Molina next challenges the ALJ’s conclusion that Molina’s allegations regarding the severity of her symptoms and their effect on her ability to work were not credible. Molina argues that the ALJ erred by basing her conclusion on activities of daily living that were not transferable to a work setting, and in considering Molina’s failure to make any efforts to seek counseling in 2003, 2004, or 2005. Again, we disagree.
In assessing the credibility of a claimant’s testimony regarding subjective pain or the intensity of symptoms, the ALJ engages in a two-step analysis. Vasquez v. Astrue, 572 F.3d 586, 591 (9th Cir.2009). First, the ALJ must determine whether there is “ ‘objective medical evidence of an underlying impairment which could reasonably be expected to produce the pain or other symptoms alleged.’” Id. (quoting Lingenfelter v. Astrue, 504 F.3d 1028, 1036 (9th Cir.2007)). If the claimant has presented such evidence, and there is no evidence of malingering, then the ALJ must give “ ‘specific, clear and convincing reasons’ ” in order to reject the claimant’s testimony about the severity of the symptoms. Id. (quoting Lingenfelter, 504 F.3d at 1036). At the same time, the ALJ is not “required to believe every allegation of disabling pain, or else disability benefits would be available for the asking, a result plainly contrary to 42 U.S.C. § 423(d)(5)(A).” Fair v. Bowen, 885 F.2d 597, 603 (9th Cir.1989). In evaluating the claimant’s testimony, the ALJ may use “ ‘ordinary techniques of credibility evaluation.’ ” Turner, 613 F.3d at 1224 n. 3 (quoting Smolen, 80 F.3d at 1284). For instance, the ALJ may consider inconsistencies either in the claimant’s testimony or between the testimony and the claimant’s conduct, id.; “ ‘unexplained or inadequately explained failure to seek treatment or to follow a prescribed course of treatment,’ ” Tommasetti, 533 F.3d at 1039 (quoting Smolen, 80 F.3d at 1284); and “whether the claimant engages in daily activities inconsistent with the alleged symptoms,” Lingenfelter, 504 F.3d at 1040. While a claimant need not “ ‘vegetate in a dark room’ ” in order to be eligible for *1113benefits, Cooper v. Bowen, 815 F.2d 557, 561 (9th Cir.1987) (quoting Smith v. Califano, 687 F.2d 968, 971 (3d Cir.1981)), the ALJ may discredit a claimant’s testimony when the claimant reports participation in everyday activities indicating capacities that are transferable to a work setting, see Morgan v. Comm’r Soc. Sec. Admin., 169 F.3d 595, 600 (9th Cir.1999); Fair, 885 F.2d at 603. Even where those activities suggest some difficulty functioning, they may be grounds for discrediting the claimant’s testimony to the extent that they contradict claims of a totally debilitating impairment. See Turner, 613 F.3d at 1225; Valentine, 574 F.3d at 693.
Here, the ALJ found that Molina’s claimed inability to tolerate even minimal human interaction was inconsistent with her daily activities throughout the disability period. This finding is supported by substantial evidence in the record. The ALJ could reasonably conclude that Molina’s activities, including walking her two grandchildren to and from school, attending church, shopping, and taking walks, undermined her claims that she was incapable of being around people without suffering from debilitating panic attacks. It was also reasonable for the ALJ to conclude that Molina’s daily activities involved a degree of human interaction that was analogous to that required by her past relevant work, which involved working at a machine by herself in a large room that was occupied by only four or five other people, with whom she did not generally need to interact.
Moreover, the ALJ supported her conclusion that Molina was not credible on the additional grounds that Molina’s allegations were undermined by her demeanor and presentation as described by Dr. Yost, and inconsistent with other medical evidence in the record. The first finding is supported by Dr. Yost’s report that Molina was alert and oriented, maintained good eye contact, did not appear excessively anxious, spoke coherently and fluently, smiled appropriately, and was congenial. The second is supported by the conclusions of both Dr. Yost and the state examining physician that Molina’s anxiety disorder was not severe and that she was able to control it with Lorazepam and other self-calming measures. Because the ALJ’s adverse credibility determination was supported by specific, clear, and convincing reasons, we uphold it.
We also reject Molina’s claim that the ALJ erred in relying on Molina’s failure to seek or follow prescribed treatment. First, the ALJ did not expressly place any weight on this factor in discounting Molina’s credibility. But to the extent the ALJ implicitly considered Molina’s failure to follow Wheelwright’s advice that she seek counseling, the ALJ did not err. We have long held that, in assessing a claimant’s credibility, the ALJ may properly rely on “ ‘unexplained or inadequately explained failure to seek treatment or to follow a prescribed course of treatment.’ ” Tommasetti, 533 F.3d at 1039 (quoting Smolen, 80 F.3d at 1284); Fair, 885 F.2d at 603. According to agency rules, “the individual’s statements may be less credible if the level or frequency of treatment is inconsistent with the level of complaints, or if the medical reports or records show that the individual is not following the treatment as prescribed and there are no good reasons for this failure.” SSR 96-7p.5 Moreover, a claimant’s failure to as*1114sert a good reason for not seeking treatment, “or a finding by the ALJ that the proffered reason is not believable, can cast doubt on the sincerity of the claimant’s pain testimony.” Fair, 885 F.2d at 603. Here, the record was filled with evidence that despite Wheelwright’s repeated efforts to persuade Molina to seek psychiatric treatment for her anxiety disorder, she failed to do so until after she applied for disability benefits. Although Molina provided reasons for resisting treatment, there was no medical evidence that Molina’s resistance was attributable to her mental impairment rather than her own personal preference, and it was reasonable for the ALJ to conclude that the “level or frequency of treatment [was] inconsistent with the level of complaints.” SSR 96-7p. Molina’s reliance on Pate-Fires v. Astrue, 564 F.3d 935, 945 (8th Cir.2009), is misplaced, because in that case, the “evidence overwhelmingly demonstrate[d] Pate-Fires’s noncompliance was attributable to her mental illness,” id. at 946, whereas there was no similar evidence here.6
y
Finally, Molina argues that the ALJ erred by failing to properly discuss the testimony of Molina’s family members.
A
Lay testimony as to a claimant’s symptoms or how an impairment affects the claimant’s ability to work is competent evidence that the ALJ must take into account. Nguyen v. Chater, 100 F.3d 1462, 1467 (9th Cir.1996); Dodrill v. Shalala, 12 F.3d 915, 919 (9th Cir.1993). We have held that competent lay witness testimony “cannot be disregarded without comment,” Nguyen, 100 F.3d at 1467, and that in order to discount competent lay witness testimony, the ALJ “must give reasons that are germane to each witness,” Dodrill, 12 F.3d at 919. We have not, however, required the ALJ to discuss every witness’s testimony on a individualized, witness-by-witness basis. Rather, if the ALJ gives germane reasons for rejecting testimony by one witness, the ALJ need only point to those reasons when rejecting similar testimony by a different witness. See Valentine, 574 F.3d at 694 (holding that because “the ALJ provided clear and convincing reasons for rejecting [the claimant’s] own subjective complaints, and because [the lay witness’s] testimony was similar to such complaints, it follows that the ALJ also gave germane reasons for rejecting [the lay witness’s] testimony”). The applicable regulations are in accord; they require the ALJ to consider testimony from family and friends submitted on behalf of the claimant, see 20 C.F.R. §§ 404.1529(c)(3), 404.1545(a)(3), but do not require the ALJ to provide express reasons for rejecting testimony from each lay witness, see id.; see also SSR 06-03p (recognizing that “there is a distinction between what an adjudicator must consider and what the adjudicator must explain in the disability determination or decision”).
Here, the ALJ stated that the rationale for her credibility determination “include[d] reference to ... the third party statements submitted in support of the claimant.” This statement establishes that the ALJ reviewed the lay witness testimo*1115ny in the record, but it does not provide a reason for discounting the testimony. The ALJ gave reasons for rejecting Molina’s testimony regarding her symptoms that were equally relevant to the similar testimony of the lay witnesses, and that would support a finding that the lay testimony was similarly not credible. But under our rule that lay witness testimony “cannot be disregarded without comment,” Nguyen, 100 F.3d at 1467, the ALJ erred in failing to explain her reasons for disregarding the lay witness testimony, either individually or in the aggregate. Nevertheless, as discussed below, that error was harmless.
B
We have long recognized that harmless error principles apply in the Social Security Act context. Stout, 454 F.3d at 1054 (collecting cases). We have, for example, deemed errors harmless where the ALJ misstated the facts about the claimant, but we were able to conclude from the record that the ALJ would have reached the same result absent the error. See Curry v. Sullivan, 925 F.2d 1127, 1131 (9th Cir.1990) (holding that the ALJ’s error in stating that the claimant was 50 years old and had a G.E.D. was harmless given that the claimant was in the “closely approaching advanced age” category regardless of the eiTor and was able to communicate in the English language). We have also deemed legal errors harmless where it was clear they did not alter the ALJ’s decision. For instance, in Brawner v. Secretary of Health & Human Services, we held that an ALJ’s error in classifying the claimant’s past relevant work as “light” was inconsequential where the record supported the ALJ’s determination that the claimant was able to perform other light work and therefore was not disabled. 839 F.2d 432, 434 (9th Cir.1988) (per curiam). And in Matthews v. Shalala, we held that even if the ALJ erred in fading to mention the claimant’s inability to stay in one position to the vocational expert, such error was harmless because the claimant had not shown that he was unable to return to his previous job as a receiving clerk and inspector, which, the record demonstrated, involved a combination of sitting and standing. 10 F.3d 678, 681 (9th Cir.1993).
In addition, several of our cases have held that an ALJ’s error was harmless where the ALJ provided one or more invalid reasons for disbelieving a claimant’s testimony, but also provided valid reasons that were supported by the record. See Bray, 554 F.3d at 1227; Carmickle v. Comm’r, Soc. Sec. Admin., 533 F.3d 1155, 1162-63 (9th Cir.2008); Batson v. Comm’r of Soc. Sec. Admin., 359 F.3d 1190, 1195— 97 (9th Cir.2004). In this context, we have said that an error is harmless so long as there remains substantial evidence supporting the ALJ’s decision and the error “does not negate the validity of the ALJ’s ultimate conclusion.” Batson, 359 F.3d at 1197; see also Carmickle, 533 F.3d at 1162.
Although we have expressed different formulations of the harmless error rule depending on the facts of the case and the error at issue, we have adhered to the general principle that an ALJ’s error is harmless where it is “inconsequential to the ultimate nondisability determination.” Carmickle, 533 F.3d at 1162; Tommasetti, 533 F.3d at 1038; Robbins v. Soc. Sec. Admin., 466 F.3d 880, 885 (9th Cir.2006); Stout, 454 F.3d at 1055. In other words, in each case we look at the record as a whole to determine whether the error alters the outcome of the case.
We applied these harmless error principles in Stout, where the ALJ had erred by neglecting to comment on competent lay witness testimony. Stout, 454 F.3d at 1054. Specifically, the claimant’s sister and brother-in-law (who worked with the claimant for 15 years) testified that the *1116claimant was unable to accomplish the simplest of tasks without near-constant supervision. Id. at 1053. We emphasized that this testimony about how the claimant’s mental impairments affected his ability to work was “uncontradicted” and “consistent with medical evidence.” Id. In holding that the ALJ’s silent disregard of the testimony was not harmless, Stout explained: “The [vocational expert] specifically opined that constant supervision is unacceptable in competitive employment; yet, the ALJ articulated no reasons for dismissing the uncontradicted lay testimony indicating Stout needed such supervision to perform even simple tasks.” Id. at 1056. Because there was nothing in the ALJ’s opinion addressing the limitations described by the lay witnesses, we were left “with nothing to review to determine whether the error materially impacted the ALJ’s ultimate decision.” Id. We explained that “where the ALJ’s error lies in a failure to properly discuss competent lay testimony favorable to the claimant, a reviewing court cannot consider the error harmless unless it can confidently conclude that no reasonable ALJ, when fully crediting the testimony, could have reached a different disability determination.” Id. Applying this rule, we determined that the ALJ’s failure to consider the testimony at issue could not be deemed “inconsequential to the ultimate nondisability determination,” id. at 1055, because the testimony identified limitations not considered by the ALJ, was uncontradicted by anything in the record, and was highly probative of Stout’s inability to work in a competitive environment.
Our subsequent decision in Robbins applied the Stout rule that the ALJ may not disregard lay witness testimony about the claimant’s limitations if the ALJ has not validly rejected those limitations and their existence would alter the ultimate nondisability determination. See Robbins, 466 F.3d at 885. In Robbins, we held that all of the ALJ’s reasons for rejecting the claimant’s allegations of subjective pain were legally invalid. Id. at 883-85. We then addressed the ALJ’s failure to discuss the corroborative lay witness testimony of the claimant’s son. Id. at 885. As in Stout, the lay witness testimony was unrebutted by anything in the record; in fact, it was substantially similar to the testimony of the claimant’s daughter, which the ALJ found “generally credible.” Robbins, 466 F.3d at 890 (O’Scannlain, J., dissenting). Robbins reasoned that “if credited, the testimony of Robbins’s son adds substantial weight not only to Robbins’s claim, but also to the testimony of Robbins’s wife and daughter, which support Robbins’s claim.” Id. at 885 (maj.op.). In this situation, the court held, “[b]eeause the ALJ did not make a legally sufficient adverse credibility finding with regard to Robbins’s own testimony, we cannot say with respect to [the son’s] testimony that ‘no reasonable ALJ, when fully crediting the testimony, could have reached a different disability determination.’ ” Id. (quoting Stout, 454 F.3d at 1056).7 In other words, given that the ALJ had not validly rejected the claimant’s excess pain testimony, the lay witness’s corroborative testimony (if credited by the ALJ) would alter the ALJ’s disability decision. Therefore, the ALJ’s failure to mention the testimony was not harmless. By the same token, Robbins suggested that the ALJ’s failure to discuss the son’s testimony would have been harmless if the ALJ had provided legally sufficient reasons for rejecting the claimant’s own testimony. See id.; see also Valentine, 574 F.3d at 694 (indicating that it is not harmful error for the ALJ to fail to discuss *1117lay witness testimony where the ALJ has provided sufficient reasons for rejecting similar testimony).
In claiming that the ALJ’s error here is not harmless, Molina relies heavily on Stout’s statement that “where the ALJ’s error lies in a failure to properly discuss competent lay testimony favorable to the claimant, a reviewing court cannot consider the error harmless unless it can confidently conclude that no reasonable ALJ, when fully crediting the testimony, could have reached a different disability determination.” Stout, 454 F.3d at 1056. Plucking this language out of context, she claims that an ALJ’s failure to give individualized reasons for rejecting a lay witness’s testimony that would be material standing alone is per se prejudicial, even if the ALJ gave well-supported reasons for rejecting similar testimony. Under this interpretation of Stout, if ten lay witnesses submitted statements that the claimant is unable to lift more than ten pounds and thus cannot return to her previous work, and the ALJ properly rejected the testimony of nine of them because it was belied by the medical evidence in the record, the ALJ’s failure to comment on the tenth witness’s testimony would require reversal.8
We disagree. Stout made clear that its holding was “consistent with our prior harmless error cases” in the Social Security context. Id. Interpreting Stout as creating a rule that the ALJ’s failure to expressly reject any facially material lay witness testimony is per se prejudicial would run afoul of our settled rule that we will not reverse for errors that are “inconsequential to the ultimate nondisability determination.” See Carmickle, 533 F.3d at 1162 (“Likewise, in Stout, after surveying our precedent applying harmless error in social security cases, we concluded that ‘in each case, the ALJ’s error ... was inconsequential to the ultimate nondisability determination.’ ” (quoting Stout, 454 F.3d at 1055)) (emphasis omitted); Tommasetti, 533 F.3d at 1038; Robbins, 466 F.3d at 885. Where lay witness testimony does not describe any limitations not already described by the claimant, and the ALJ’s well-supported reasons for rejecting the claimant’s testimony apply equally well to the lay witness testimony, it would be inconsistent with our prior harmless error precedent to deem the ALJ’s failure to discuss the lay witness testimony to be prejudicial per se. See also Valentine, 574 F.3d at 694 (holding that if an ALJ gave germane reasons for rejecting the claimant’s testimony, those reasons are equally germane to similar testimony by a lay witness.); Lewis, 236 F.3d at 512 (stating that the ALJ “noted arguably germane reasons for dismissing the [lay] testimony, even if he did not clearly link his determination to those reasons”).
Molina’s interpretation of Stout would also run afoul of the Supreme Court’s recent decision in Shinseki v. Sanders. See Sanders, 556 U.S. at 409, 129 S.Ct. 1696. In Sanders, the Supreme Court rejected a test developed by the Federal Circuit for determining whether the Department of Veterans Affairs’s failure to give the dis*1118ability claimant certain statutorily required notice was harmless. Id. at 399, 129 S.Ct. 1696. The Federal Circuit had held that the Veteran’s Court should presume any such error is prejudicial and reverse unless the Department could make a specified showing.9 Id. at 403-04, 407, 129 S.Ct. 1696. According to the Supreme Court, this test conflicted with the Veteran Court’s statutory duty to “take due account of the rule of prejudicial error,” id. at 406, 129 S.Ct. 1696 (quoting 38 U.S.C. § 7261(b)(2)), which the Supreme Court held “requires the Veterans Court to apply the same kind of ‘harmless-error’ rule that courts ordinarily apply in civil cases,” id. In other words, courts must review cases “ ‘without regard to errors’ that do not affect the parties’ ‘substantial rights.’ ” Id. at 407, 129 S.Ct. 1696 (quoting 28 U.S.C. § 2111, which codifies the federal harmless error rule for civil cases).
In expounding on this principle, Sanders focused on the error of relying on “mandatory presumptions and rigid rules,” reasoning that such reliance frustrates Congress’s express preference for determining harmlessness by “case-specific application of judgment, based upon examination of the record.” Id. at 407, 129 S.Ct. 1696 (citing Kotteakos v. United States, 328 U.S. 750, 759-60, 66 S.Ct. 1239, 90 L.Ed. 1557 (1946)). The use of presumptions, the Court explained, “exhibits] the very characteristics that Congress sought to discourage,” because it prevents the court “from resting its conclusion on the facts and circumstances of the particular ease,” and therefore requires the reviewing court to find the error prejudicial “even if that court, having read the entire record, conscientiously concludes the contrary.” Id. at 408, 129 S.Ct. 1696. The Court held that the Federal Circuit’s use of a presumption “increase[d] the likelihood of reversal in cases where, in fact, the error is harmless.” Id. at 409, 129 S.Ct. 1696. Such a presumption only “encourages abuse of the judicial process and diminishes the public’s confidence in the fair and effective operation of the judicial system.” Id. While the Court acknowledged that “courts may sometimes make empirically based generalizations about what kinds of errors are likely, as a factual matter, to prove harmful,” id. at 411, 129 S.Ct. 1696, it emphasized that courts should “hesita[te] to generalize too broadly about particular kinds of errors when the specific factual circumstances in which the error arises may well make all the difference,” id. at 412, 129 S.Ct. 1696. Such generalizations are better left to the lower court that “sees sufficient case-specific raw material” to enable it to draw empirical conclusions. Id. at 412, 129 S.Ct. 1696.
Molina’s interpretation of Stout as creating a rule that an ALJ’s failure to give individual reasons for rejecting a lay witness’s material testimony is per se prejudicial cannot be reconciled with the reasoning in Sanders. First, like the Federal Circuit’s presumption, it would prevent the reviewing court from making a “case-specific application of judgment, based upon examination of the record,” id. at 398, 129 S.Ct. 1696, and would require the reviewing court to reverse in many cases where it is “obvious from the record ... that the error made no difference,” id. at 407, 129 S.Ct. 1696. Second, Molina’s interpretation of Stout is not an empirically-based generalization “about what kinds of errors are likely, as a factual matter, to prove harmful.” Id. at 411, 129 S.Ct. 1696. Molina has not suggested any basis for Con-*1119eluding that an ALJ’s failure to discuss a lay witness’s testimony expressly is likely to affect the outcome in situations where the testimony is similar to other testimony that the ALJ validly discounted, or where the testimony is contradicted by more reliable medical evidence that the ALJ credited. Moreover, as the Supreme Court pointed out, we are not well situated to make such generalizations, since we are removed from the “case-specific raw material.” Id.
Finally, if we interpreted Stout the way Molina asks us to, we would ignore the Supreme Court’s instruction that we are “to apply the same kind of ‘harmless-error’ rule that courts ordinarily apply in civil cases.” Id. at 406, 129 S.Ct. 1696.10 That harmless error rule, as codified in 28 U.S.C. § 2111, requires us to “give judgment after an examination of the record without regard to errors or defects which do not affect the substantial rights of the parties.” 28 U.S.C. § 2111; see also Sanders, 556 U.S. at 407, 129 S.Ct. 1696. Interpreting § 2111, we have held that a district court’s erroneous exclusion of evidence does not warrant reversal unless the error “more probably than not tainted the verdict.” Engquist v. Or. Dep’t of Agric., 478 F.3d 985, 1009 (9th Cir.2007), aff'd 553 U.S. 591, 128 S.Ct. 2146, 170 L.Ed.2d 975 (2008); see also Haddad v. Lockheed Cal. Corp., 720 F.2d 1454, 1459 (9th Cir.1983) (“Just as the verdict in a civil case need only be more probably than not true, so an error in a civil trial need only be more probably than not harmless.”).11 We have deemed several factors important in determining harmlessness, including whether the evidence at issue was “cumulative of other competent testimony,” Haddad, 720 F.2d at 1460; see also id. at 1456 (“Since this evidence is cumulative of other evidence in the record and the record contains no evidence to the contrary, we presume that the improper admission of this testimony had no effect on the court’s decision.”); Obrey v. Johnson, 400 F.3d 691, 701 (9th Cir.2005) (concluding that the district court’s erroneous exclusion of relevant witness testimony was not harmless because the evidence “was not merely tangential or cumulative”); Bank of the West v. Commercial Credit Fin. Servs., 852 F.2d 1162, 1166 n. 2 (9th Cir.1988) (concluding that the district court’s erroneous exclusion of a witness’s declaration was harmless because the declaration was “cumulative of other evidence”), and the “overall strength of [the] case” against the appealing party, see Boyd v. City & Cnty. of S.F., 576 F.3d 938, 949 (9th Cir.2009); United States v. Morales, 108 F.3d 1031, 1040 (9th Cir.1997) (en banc). Molina’s interpretation of Stout would conflict with the harmless error standard we apply in civil cases because it would prevent us from considering these crucial factors. If the ALJ’s failure to address lay witness testimony *1120that is material on its face is always prejudicial, it would be irrelevant whether such testimony is cumulative of other testimony that the ALJ properly rejected. Cf. Robbins, 466 F.3d at 885 (finding reversible error “[bjecause the ALJ did not make a legally sufficient adverse credibility finding with regard to” similar testimony from the claimant himself); Valentine, 574 F.3d at 694. It would be equally irrelevant whether substantial evidence credited by the ALJ contradicts the lay witness testimony on material points. Cf. Stout, 454 F.3d at 1053-57 (finding reversible error where the lay witness testimony was “uncontradicted” and “consistent with medical evidence”).
We also note Sanders’s warning that the harmless error test in civil cases should not be as stringent as that in criminal cases. As the Supreme Court explained, “the fact that the Government must prove its case beyond a reasonable doubt [in criminal cases] justifies a rule that makes it more difficult for the reviewing court to find that an error did not affect the outcome of a case. But in the ordinary civil case that is not so.” Sanders, 556 U.S. at 410-11, 129 S.Ct. 1696 (internal citation omitted); see also Bank of Nova Scotia v. United States, 487 U.S. 250, 255-56, 108 S.Ct. 2369, 101 L.Ed.2d 228 (1988) (“It would be inappropriate to devise a rule permitting federal courts to deal more sternly with nonconstitutional harmless errors than with constitutional errors that are likewise harmless.”). But Molina interprets Stout as setting a higher standard than is required to protect a criminal defendant’s constitutional rights. Whereas in criminal cases, “an otherwise valid conviction should not be set aside if the reviewing court may confidently say, on the whole record, that the constitutional error was harmless beyond a reasonable doubt,” Delaware v. Van Arsdall, 475 U.S. 673, 681, 106 S.Ct. 1431, 89 L.Ed.2d 674 (1986) (emphasis added), Molina’s interpretation would prevent the reviewing court from considering whether the ALJ had given reasons for rejecting the undiscussed lay •witness testimony that were supported by substantial evidence in the context of the record as a whole. And whereas in criminal cases, the reviewing court is directed to consider factors such as “whether the testimony was cumulative, the presence or absence of evidence corroborating or contradicting the testimony of the witness on material points ..., and ... the over-all strength of the prosecution’s case,” id. at 684, 106 S.Ct. 1431, Molina’s interpretation would, as discussed, prevent us from considering these important factors at all.
The dissent suggests that Molina’s proposed rule is appropriate because a reviewing court cannot make independent credibility determinations. Dis. op. at 1122. Therefore, the dissent argues, if the ALJ fails to make an individual determination of the credibility of each lay witness, we must presume that the lay witness is credible and give full effect to the witness’s testimony.12 We do not disagree with the general proposition that it is the *1121ALJ’s prerogative to determine the credibility of witnesses. Our role as a reviewing court is limited in this regard, both because the administrative judge is generally in a unique position to observe the witnesses’ demeanor and conduct firsthand, see Universal Camera Corp. v. NLRB, 340 U.S. 474, 496, 71 S.Ct. 456, 95 L.Ed. 456 (1951),13 and because we may not uphold an agency’s decision on a ground not actually relied on by the agency, see SEC v. Chenery Corp., 332 U.S. 194, 196, 67 S.Ct. 1575, 91 L.Ed. 1995 (1947). But for the same reasons, if an ALJ has provided well-supported grounds for rejecting testimony regarding specified limitations, we cannot ignore the ALJ’s reasoning and reverse the agency merely because the ALJ did not expressly discredit each witness who described the same limitations. Further, where the ALJ rejects a witness’s testimony without providing germane reasons, but has already provided germane reasons for rejecting similar testimony, we cannot reverse the agency merely because the ALJ did not “clearly link his determination to those reasons.” Lewis, 236 F.3d at 512. Even when an agency “explains its decision with ‘less than ideal clarity,’ ” we must uphold it “if the agency’s path may reasonably be discerned.” Alaska Dep’t of Envtl. Conservation v. EPA, 540 U.S. 461, 497, 124 S.Ct. 983, 157 L.Ed.2d 967 (2004) (quoting Bowman Transp., Inc. v. Arkansas-Best Freight Sys., 419 U.S. 281, 286, 95 S.Ct. 438, 42 L.Ed.2d 447 (1974)). Giving full effect to the ALJ’s reasoning does not require us to consider whether the undiscussed lay witness was credible or not credible. Rather, it is Molina’s proposed interpretation of Stout that would require the reviewing court to make a credibility determination where the ALJ had not. Specifically, under Molina’s proposed rule, the reviewing court must deem the undiscussed lay witness testimony to be both credible and disproportionately important, so as to compel the conclusion that the ALJ’s decision was not supported by substantial evidence. By contrast, our decision today accords the proper deference to the agency factfinder by applying the “highly deferential standard” that requires us to affirm the ALJ’s decision if supported by “ ‘such relevant evidence as a reasonable mind might accept as adequate to support a conclusion.’ ” Valentine, 574 F.3d at 690 (quoting Desrosiers, 846 F.2d at 576). A reviewing court’s refusal to consider whether the ALJ’s reasoning applies to undiscussed lay witness testimony is contrary not only to our case law holding that errors are harmless if they are “inconsequential to the ultimate nondisability determination,” see, e.g., Carmickle, 533 F.3d at 1162, but also to the long-settled rule that we will not set aside the denial of a disability claim unless “the Secretary’s findings are not supported by substantial evidence in the record as a whole,” Stone, 761 F.2d at 531 (emphasis added).
As we held in another recent social security case, Sanders makes it “quite clear that no presumptions operate,” and we must analyze harmlessness “in light of the circumstances of the case.” McLeod v. Astrue, 640 F.3d 881, 888 (9th Cir.2011). Accordingly, we reject Molina’s interpretation of Stout as creating a per se rule of prejudice when the ALJ fails to discuss lay witness testimony that is material if considered by itself, and instead apply Stout according to our generally applicable harmless error principles. In reaching this conclusion, we join the Eighth Cir*1122cuit’s well reasoned determination that an ALJ’s failure to comment upon lay witness testimony is harmless where “the same evidence that the ALJ referred to in discrediting [the claimant’s] claims also discredits [the lay witness’s] claims.” Buckner v. Astrue, 646 F.3d 549, 560 (8th Cir.2011).
C
Turning to the case at hand, we must consider whether the ALJ’s failure to discuss the testimony from Molina’s family members was “inconsequential to the ultimate nondisability determination” in the context of the record as a whole. See Carmickle, 533 F.3d at 1162; Tommasetti 533 F.3d at 1038; Robbins, 466 F.3d at 885; Stout, 454 F.3d at 1055. Here, the ALJ failed to explain her reasons for rejecting the lay witnesses’ testimony. That testimony, however, did not describe any limitations beyond those Molina herself described,14 which the ALJ discussed at length and rejected based on well-supported, clear and convincing reasons. Specifically, the ALJ determined that Molina’s claim that her anxiety disorder made her unable to work was contradicted by her demeanor and presentation, by Dr. Yost’s evaluation, and by her own testimony about her daily activities throughout the alleged disability period. Because the ALJ had validly rejected all the limitations described by the lay witnesses in discussing Molina’s testimony, we are confident that the ALJ’s failure to give specific witness-by-witness reasons for rejecting the lay testimony did not alter the ultimate nondisability determination. Accordingly, the ALJ’s error was harmless.
VI
The ALJ did not err in weighing the evidence as she did or in finding that Molina’s testimony regarding the severity of her impairment was not credible. Although the ALJ erred in failing to give germane reasons for rejecting the lay witness testimony, such error was harmless given that the lay testimony described the same limitations as Molina’s own testimony, and the ALJ’s reasons for rejecting Molina’s testimony apply with equal force to the lay testimony. Applying the principles set forth in our social security eases, as well as in Sanders, we uphold the ALJ’s decision as supported by substantial evidence.
AFFIRMED.
. On appeal, Molina does not challenge the ALJ’s findings that her physical impairments do not restrict her ability to work, so we do not discuss them here.
. These are limited to licensed or certified psychologists, licensed optometrists, licensed podiatrists, and qualified speech-language pathologists. 20 C.F.R. § 404.1513(a).
. In holding that a nurse practitioner could be an acceptable medical source, Gomez relied in part on language in 20 C.F.R. § 416.913(a)(6), which stated that "[a] report of an interdisciplinary team that contains the evaluation and signature of an acceptable medical source is also considered acceptable medical evidence.” See Gomez, 74 F.3d at 971. This regulatory section has since been repealed. We need not, however, address Gomez's continued vitality because Wainwright acted alone.
. Some language in the ALJ’s opinion suggests the ALJ may have considered Wheelwright to be a medically acceptable source; specifically, the ALJ discussed the deference generally given to treating sources under 20 C.F.R. § 404.1527(d)(2). Even assuming the ALJ erred in this regard, the error was harmless because the ALJ gave specific, legitimate reasons based on substantial evidence in the record for discounting Wheelwright’s testimony. Valentine, 574 F.3d at 692.
. Social Security Rulings (SSRs) "do not carry the ‘force of law,' but they are binding on ALJs nonetheless.” Bray v. Comm’r Soc. Sec. Admin., 554 F.3d 1219, 1224 (9th Cir.2009). They " 'reflect the official interpretation of the [SSA] and are entitled to some deference as long as they are consistent with the Social Security Act and regulations.’ ” Id. (alteration in original) (quoting Avenetti v. Barnhart, 456 F.3d 1122, 1124 (9th Cir.2006)).
. We also reject Molina's argument that the ALJ erred by failing to comply with SSR 82-59, which provides that an ALJ may deny benefits to a claimant who has a disability if the claimant unjustifiably fails to follow prescribed treatment that is "clearly expected to restore capacity to engage in any [substantial gainful activity].” This rule is not applicable here, because the ALJ determined that Molina was not disabled, and Molina’s failure to seek treatment (to the extent the ALJ considered it at all) was merely a factor in the ALJ's credibility determination. See Roberts v. Shalala, 66 F.3d 179, 183 (9th Cir.1995).
. Although Robbins held that the ALJ’s failure to discuss the son’s testimony was reversible error, the ALJ’s decision would have required reversal in any event due to the ALJ's failure to give legally sufficient reasons for rejecting the claimant’s own testimony.
. According to the dissent, Molina's interpretation of Stout does not create a per se rule because “we still must decide whether the testimony affected the disability determination." Dis. op. at 1123. We disagree. Claimants do not generally submit lay witness statements unless they support their disability claim, and thus such statements will naturally affect the disability determination if considered on their own. Because Molina's proposed interpretation of Stout would preclude the reviewing court from considering undiscussed lay witness statements in the context of the record as a whole, it would effectively create a per se rule of prejudice whenever the ALJ fails to give an individualized reason for rejecting such statements.
. Specifically, the Veterans Court was required to find the error harmful unless the Department could demonstrate "(1) that the claimant's ‘actual knowledge’ cured the defect or (2) that the claimant could not have received a benefit as a matter of law.” Id. at 407, 129 S.Ct. 1696.
. Although Sanders interpreted a different harmless error statute specific to the Veterans Administration, see 38 U.S.C. § 7261(b)(2), we have already held that the same harmless error standard applies to Social Security cases because § 7261 merely "imported] the Administrative Procedure Act rule and the general rule for federal cases.” McLeod v. Astrue, 640 F.3d 881, 887 (9th Cir.2011).
. In Obrey v. Johnson, we held that "[t]he party benefitting from the error has the burden of persuasion, and 'in cases of equipose, we reverse.' " 400 F.3d 691, 701 (9th Cir.2005) (quoting United States v. Seschillie, 310 F.3d 1208, 1214-15 (9th Cir.2002)). We do not for present purposes need to decide whether this holding survives Sanders, which held that in ordinary civil cases “the party that ‘seeks to have a judgment set aside because of an erroneous ruling carries the burden of showing that prejudice resulted.’ ” Sanders, 556 U.S. at 409, 129 S.Ct. 1696 (quoting Palmer v. Hoffman, 318 U.S. 109, 116, 63 S.Ct. 477, 87 L.Ed. 645 (1943)). Regardless of which party bears the burden of persuasion, we conclude that the error was harmless in this case.
. The dissent analogizes to cases in the immigration context holding that we presume that an asylum petitioner is credible if the immigration judge did not make an explicit adverse credibility finding. Dis. op. at 1123 (citing Abovian v. INS, 219 F.3d 972, 978 (9th Cir.), as amended by 228 F.3d 1127, 1127 (9th Cir.2000)); see also Kalubi v. Ashcroft, 364 F.3d 1134, 1137-38 (9th Cir.2004). But see 8 U.S.C. § 1158(b)(1)(B)(iii) (2008) (providing that "if no adverse credibility determination is explicitly made, the applicant or witness shall have a rebuttable presumption of credibility on appeal” (emphasis added)); Su Hwa She v. Holder, 629 F.3d 958, 964 n. 5 (9th Cir.2010). In our view, these immigration cases are unhelpful because they do not address the issue before us, namely, when an administrative judge’s failure to expressly comment on a witness’s testimony is prejudicial.
. Here, of course, the lay witness testimony was in the form of written statements, not live testimony given under oath, so firsthand observations are irrelevant to the credibility determination.
. Contrary to the dissent, dis. op. at 1123, the letters from Molina's family members described the same limitations as Molina described in her function reports and testimony before the ALJ, including her reluctance to travel, her inability to finish her food at restaurants, and her physical symptoms during a panic attack. See supra pp. 1109-10.