AMENDED OPINIONWALLACE, Senior Circuit Judge

Eric G. Slepian (argued), Phoenix, AZ, for Plaintiff–Appellant.

Jessica Milano (argued) and Matthew M. Linton, Special Assistant United States Attorneys, and John Jay Lee, Regional Chief Counsel, Region VIII, Social Security Administration Office of the General Counsel, Denver, CO; John S. Leonardo, United States Attorney, Michael Johns, Assistant United States Attorney, Denver, CO, for Defendant–Appellee.

Appeal from the United States District Court for the District of Arizona, Frederick J. Martone, Senior District Judge, Presiding. D.C. No. 2:11–cv–02573–FJM.

Before: J. CLIFFORD WALLACE, MILAN D. SMITH, JR., and PAUL J. WATFORD, Circuit Judges.

ORDER

The court's opinion filed August 4, 2015, and appearing at 798 F.3d 749 (9th Cir.2015), is hereby amended. An amended opinion is filed herewith.

With this amended opinion, the panel has voted to deny the petition for panel rehearing. That petition is therefore DENIED. No further petitions for rehearing or rehearing en banc may be filed.

AMENDED OPINION

WALLACE, Senior Circuit Judge:

When an Administrative Law Judge (ALJ) determines that a claimant for Social Security benefits is not malingering and has provided objective medical evidence of an underlying impairment which might reasonably produce the pain or other symptoms she alleges, the ALJ may reject the claimant's testimony about the severity of those symptoms only by providing specific, clear, and convincing reasons for doing so. We hold that an ALJ does not provide specific, clear, and convincing reasons for rejecting a claimant's testimony by simply reciting the medical evidence in support of his or her residual functional capacity determination. To ensure that our review of the ALJ's credibility determination is meaningful, and that the claimant's testimony is not rejected arbitrarily, we require the ALJ to specify which testimony she finds not credible, and then provide clear and convincing reasons, supported by evidence in the record, to support that credibility determination.

Here, the ALJ found generally that the claimant's testimony was not credible, but failed to identify which testimony she found not credible and why. We conclude, therefore, that the ALJ committed legal error. This error was not harmless because it precludes us from conducting a meaningful review of the ALJ's reasoning. We therefore vacate the district court's judgment affirming the ALJ's denial of benefits. Because we conclude that critical factual issues remain unresolved, and that further proceedings will be useful, we instruct the district court to remand this case to the ALJ for further proceedings rather than for an immediate award of benefits.

I.

Kim Brown–Hunter applied for Social Security disability benefits and supplemental security income on April 29, 2009. After the Social Security Administration denied her claims initially and again on reconsideration, Brown–Hunter timely requested and was granted a hearing before an ALJ.

A.

At the hearing, Brown–Hunter provided medical evidence that she suffered from obesity, peripheral neuropathy, lumbar degenerative disc disease, diabetes mellitus, migraine headaches, hypothyroidism, hypertension, and asthma. Brown–Hunter told the ALJ that until 2003 she had worked as a certified nursing assistant, but had to stop working because she “could no longer lift the patients or get them out of bed or in bed.” She testified that she looked for other work until about 2005, when “it just got to the point where [she] couldn't sit up or stand up or walk a distance.”

Brown–Hunter next described her then-current functional limitations. She testified that she could not drive long distances, that she could lift only about ten pounds, and that she was able to sit only for about an hour and to stand only for about forty-five minutes. She described her daytime routine as “sit[ting] in the garage for a couple hours,” which was often interrupted by the need to lie down “like four or five times for about thirty minutes” because her “feet and ... legs [would] swell up when [she was] on them too much.” In the evenings, she stated she would talk, visit, and watch television with her son and daughter, and that if she needed something from the store, her daughter would get it for her. When the ALJ asked about her ability to do housework, Brown–Hunter stated that she was able to “pick up some, but [was] not able to sweep anymore or mop or vacuum or bend down for the dishwasher.”

Brown–Hunter told the ALJ that her functional limitations were caused in part by severe back and leg pain, coupled with swelling in her lower extremities. When the ALJ asked if anything alleviated this pain and swelling, Brown–Hunter replied that elevating her feet, as prescribed by her doctors, helped with the leg and feet pain, but that “[w]ith my back I have to l [ie] down.” Brown–Hunter stated that to control the pain and swelling effectively, she needed to elevate her feet twice a week for thirty minutes to an hour, and to lie down three to four times a day for thirty minutes to an hour. Brown–Hunter also provided evidence that she was taking roughly twenty medications, including strong pain medications, such as oxycodone.

After Brown–Hunter concluded her testimony, the ALJ turned to the vocational expert. The ALJ asked the vocational expert a series of hypothetical questions to identify the occupations that existed in significant numbers in the national economy that an individual could perform, assuming such an individual had Brown–Hunter's same age, education, past work experience, and functional limitations.

The ALJ began by asking the vocational expert which occupations would be available to such individual “limited to a range of light work as ... defined under the regulations,” but with the ability to do “no more than occasional pushing and pulling with the upper and lower extremities.” The vocational expert identified a handful of possible jobs, including “light janitorial or light office cleaning,” and an unskilled, light “[a]ssembly position.” When the ALJ asked how his testimony would change if that hypothetical was modified “to add ... that the individual would be off task 15 to 20 percent of the work day,” the vocational expert replied that such individual “would not be able to sustain the work.”

The ALJ then allowed Brown–Hunter's counsel to question the vocational expert. As relevant here, counsel asked the vocational expert to assume an individual with lower extremity swelling who needed to elevate her feet thirty minutes to one hour twice a week:

Q: If we were to assume those limitations, would such an individual be able to perform Claimant's past work?

A: No.

Q: Would there be any other job they could sustain?

A: Not likely.

...

Q: And if an individual needed to rest or lie down ... two or three times a day for thirty minutes at a time, ... and it was unscheduled, would that preclude such an individual from performing Claimant's past work?

A: Yes.

Q: Would it allow for any other jobs?

A: No.

Shortly after this testimony the hearing concluded.

B.

Several weeks later, the ALJ issued a written decision denying Brown–Hunter's claims. Following the Social Security Act's five-step procedure for determining disability, see 20 C.F.R. § 416.920(a)(4), the ALJ concluded at step one that Brown–Hunter had not engaged in substantial gainful activity since the date of the alleged onset. At step two, the ALJ concluded that Brown–Hunter suffered from the “severe” impairments of “obesity, peripheral neuropathy, lumbar degenerative disc disease (DDD), diabetes mellitus (DM), and migraine headaches.” However, at step three, the ALJ concluded that those impairments did not—either in isolation or in combination—meet or equal a “listed” impairment. See 20 C.F.R. § 404, Subpart P, App'x 1. This conclusion required the ALJ to then determine Brown–Hunter's Residual Functional Capacity (RFC) in preparation for step four.

The ALJ began her discussion of the RFC with her conclusion that Brown– Hunter had the RFC “to perform light work ... except she can perform no more than occasional pushing and pulling with the upper and lower extremities.” The ALJ then acknowledged her duty to “evaluate the intensity, persistence, and limiting effects of the claimant's symptoms to determine the extent to which they limit the claimant's functioning,” and, where appropriate, to “make a finding on the credibility of the statements based on a consideration of the entire case record.” She went on to paraphrase Brown–Hunter's testimony regarding the chronology of her back pain, followed by a summary of the treatments and prescriptions for that pain, the symptoms pertaining to Brown–Hunter's diabetic peripheral neuropathy, as well as Brown–Hunter's height, weight, smoking habits, and ability “to drive short distances, sit for one hour and stand for 45 minutes.”

Following that summary, the ALJ concluded that although Brown–Hunter's “medically determinable impairments could reasonably be expected to cause some of the alleged symptoms,” her “statements concerning the intensity, persistence and limiting effects of these symptoms are not credible to the extent they are inconsistent with the above residual functional capacity assessment.”

Finally, after devoting the next eight paragraphs to summarizing the medical evidence in the record, the ALJ stated:

After careful consideration of the entire record, including the medical evidence and the testimony at the hearing, I find the functional limitations resulting from the claimant's impairments were less serious than she has alleged.... [While] the medical evidence supports a finding that the claimant's impairments ... impos[e] some restrictions, [they] did not prevent her from engaging in all work related activities.

Consequently, the ALJ proceeded to step four and concluded that Brown–Hunter's limited RFC precluded her from performing any past relevant work. But at step five, the ALJ determined that, considering Brown–Hunter's RFC, age, education, and work experience, she was able to do other work that existed in significant numbers in the national economy. Citing the vocational expert's testimony that a hypothetical individual with the “same age, education, and past relevant work experience” as Brown–Hunter could perform occupations such as “light janitorial” and “assembly,” the ALJ concluded that Brown Hunter was not disabled under the Social Security Act.

C.

The ALJ's decision became final when the Social Security Appeals Council denied review. Brown–Hunter then appealed from the ALJ's decision to the district court. She argued that the ALJ erred by “fail[ing] to provide clear and convincing reasons for rejecting her symptom testimony.” The district court rejected her argument, holding that “the ALJ gave clear and convincing reasons to support [her] determination that [Brown–Hunter's] ‘impairments were less serious than she has alleged.’ ” The district court pieced together medical evidence identified by the ALJ that it found inconsistent with “an allegation of excess pain,” and stated therefore that the ALJ “identified several inconsistencies between [Brown–Hunter's] testimony and the record,” and “gave clear and convincing reasons to support [her] determinations that [Brown–Hunter's] impairments were less serious than she has alleged.' ” The district court affirmed the ALJ's decision, and Brown–Hunter timely filed this appeal, over which we have jurisdiction pursuant to 28 U.S.C. § 1291.

II.

1 We review de novo a district court's order affirming an ALJ's denial of Social Security benefits. Hill v. Astrue, 698 F.3d 1153, 1158 (9th Cir.2012). On de novo review, we must bear in mind that a federal court's review of Social Security determinations is quite limited. “For highly fact-intensive individualized determinations like a claimant's entitlement to disability benefits, Congress ‘places a premium upon agency expertise, and, for the sake of uniformity, it is usually better to minimize the opportunity for reviewing courts to substitute their discretion for that of the agency.’ ” Treichler v. Comm'r of Soc. Sec. Admin., 775 F.3d 1090, 1098 (9th Cir.2014), quoting Consolo v. Fed. Mar. Comm'n, 383 U.S. 607, 621, 86 S.Ct. 1018, 16 L.Ed.2d 131 (1966). To ensure that we adhere to this principle, we follow three important rules in our analysis of the ALJ's decision.

234 The first is that “we leave it to the ALJ to determine credibility, resolve conflicts in the testimony, and resolve ambiguities in the record.” Treichler, 775 F.3d at 1098. The second is that we will “disturb the Commissioner's decision to deny benefits ‘only if it is not supported by substantial evidence or is based on legal error.’ ” Id., quoting Andrews v. Shalala, 53 F.3d 1035, 1039 (9th Cir.1995). The third is that “[e]ven when the ALJ commits legal error, we uphold the decision where that error is harmless,” meaning that “it is inconsequential to the ultimate nondisability determination,” or that, despite the legal error, “the agency's path may reasonably be discerned, even if the agency explains its decision with less than ideal clarity.” 775 F.3d at 1099 (internal quotation marks and citations omitted).

5 That said, “our precedents have been cautious about when harmless error should be found.” Marsh v. Colvin, 792 F.3d 1170, 1173 (9th Cir.2015). Ever mindful of our duty not to substitute our own discretion for that of the agency, we have emphasized that “the decision on disability rests with the ALJ and the Commissioner of the Social Security Administration in the first instance, not with a district court.” Id. at 1173. Thus, although we will not fault the agency merely for explaining its decision with “less than ideal clarity,” Treichler, 775 F.3d at 1099 (citation and internal quotation marks omitted), we still demand that the agency set forth the reasoning behind its decisions in a way that allows for meaningful review. A clear statement of the agency's reasoning is necessary because we can affirm the agency's decision to deny benefits only on the grounds invoked by the agency. Stout v. Comm'r, Soc. Sec. Admin., 454 F.3d 1050, 1054 (9th Cir.2006).

6 A reviewing court may not make independent findings based on the evidence before the ALJ to conclude that the ALJ's error was harmless. Id.; see also Marsh, 792 F.3d at 1172 (a district court may not find harmless error by “affirm[ing] the agency on a ground not invoked by the ALJ”). Rather, “[w]e are constrained to review the reasons the ALJ asserts.” Connett v. Barnhart, 340 F.3d 871, 874 (9th Cir.2003). If the ALJ fails to specify his or her reasons for finding claimant testimony not credible, a reviewing court will be unable to review those reasons meaningfully without improperly “substitut[ing] our conclusions for the ALJ's, or speculat[ing] as to the grounds for the ALJ's conclusions.” Treichler, 775 F.3d at 1103. Because we cannot engage in such substitution or speculation, such error will usually not be harmless.

III.

7 Where, as here, an ALJ concludes that a claimant is not malingering, and that she has provided objective medical evidence of an underlying impairment which might reasonably produce the pain or other symptoms alleged, the ALJ may “reject the claimant's testimony about the severity of her symptoms only by offering specific, clear and convincing reasons for doing so.” Lingenfelter v. Astrue, 504 F.3d 1028, 1036 (9th Cir.2007) (citation and internal quotation marks omitted). The Commissioner disputes that standard of review. Relying primarily on Bunnell v. Sullivan, 947 F.2d 341 (9th Cir.1991), the Commissioner argues that although the court reviews the ALJ's credibility finding for adequate specificity, “clear and convincing” reasons are not required. We recently rejected this argument in Burrell v. Colvin, 775 F.3d 1133, 1136–37 (9th Cir.2014) (holding that the “clear and convincing” requirement actually predated Bunnell and was not “excised” by it). Burrell therefore forecloses the Commissioner's argument.

8 A finding that a claimant's testimony is not credible “must be sufficiently specific to allow a reviewing court to conclude the adjudicator rejected the claimant's testimony on permissible grounds and did not arbitrarily discredit a claimant's testimony regarding pain.” Bunnell, 947 F.2d at 345–46 (citation and internal quotation marks omitted). “General findings are insufficient; rather, the ALJ must identify what testimony is not credible and what evidence undermines the claimant's complaints.” Reddick v. Chater, 157 F.3d 715, 722 (9th Cir.1998) (citation and internal quotation marks omitted). See also Holohan v. Massanari, 246 F.3d 1195, 1208 (9th Cir.2001) (“the ALJ must specifically identify the testimony she or he finds not to be credible and must explain what evidence undermines the testimony” (emphasis added)). The governing Social Security rulings, which “are binding on all components of the Social Security Administration, ... and are to be relied upon as precedents in adjudicating cases,” Orn v. Astrue, 495 F.3d 625, 636 (9th Cir.2007), quoting 67 Fed.Reg. at 57860, are clear on this point:

it is not sufficient for the adjudicator to make a single, conclusory statement that ‘the individual's allegations have been considered’ or that ‘the allegations are [not] credible.’ ... The determination ... must contain specific reasons for the finding on credibility, supported by the evidence in the case record, and must be sufficiently specific to make clear to ... any subsequent reviewers the weight the adjudicator gave to the individual's statements and the reasons for that weight.

Social Security Ruling 96–7P, 1996 WL 374186 at *2 (July 2, 1996) (emphasis added).

Following these principles in Treichler, for example, we held that the ALJ erred by making only a “single general statement that ‘the claimant's statements concerning the intensity, persistence and limiting effects of these symptoms are not credible to the extent they are inconsistent with the above residual functional capacity assessment,’ ” without identifying “sufficiently specific reasons” for rejecting the testimony, supported by evidence in the case record. 775 F.3d at 1102–03. The ALJ here made the identical conclusory statement and likewise failed to identify specifically which of Brown–Hunter's statements she found not credible and why. Instead, the ALJ stated only that she found, based on unspecified claimant testimony and a summary of medical evidence, that “the functional limitations from the claimant's impairments were less serious than she has alleged.”

9 We disagree with the district court that the ALJ “identified several inconsistencies between [Brown–Hunter's] testimony and the record,” and therefore “gave clear and convincing reasons to support” her non-credibility determination. Our review of the ALJ's written decision reveals that she did not specifically identify any such inconsistencies; she simply stated her non-credibility conclusion and then summarized the medical evidence supporting her RFC determination. This is not the sort of explanation or the kind of “specific reasons” we must have in order to review the ALJ's decision meaningfully, so that we may ensure that the claimant's testimony was not arbitrarily discredited. Although the inconsistencies identified by the district court could be reasonable inferences drawn from the ALJ's summary of the evidence, the credibility determination is exclusively the ALJ's to make, and ours only to review. As we have long held, “[W]e are constrained to review the reasons the ALJ asserts.” Connett v. Barnhart, 340 F.3d 871, 874 (9th Cir.2003) (emphasis added). Thus, the inconsistencies identified independently by the district court cannot provide the basis upon which we can affirm the ALJ's decision.

Indeed, “[o]ur decisions make clear that we may not take a general finding—an unspecified conflict between Claimant's testimony ... and her reports to doctors—and comb the administrative record to find specific conflicts.” Burrell, 775 F.3d at 1138. Because the ALJ failed to identify the testimony she found not credible, she did not link that testimony to the particular parts of the record supporting her non-credibility determination. This was legal error. See id. at 1139 (holding that the ALJ committed legal error because he “never connected the medical record to Claimant's testimony” nor made “a specific finding linking a lack of medical records to Claimant's testimony about the intensity of her ... pain” (emphasis added)). Even if the district court's analysis was sound, it could not overcome the error of the ALJ. That is, the error could not be corrected by the district court's statement of links between claimant testimony and certain medical evidence.

Nor was that error harmless. An error is harmless only if it is “inconsequential to the ultimate nondisability determination,” Molina v. Astrue, 674 F.3d 1104, 1115 (9th Cir.2012) (citation and internal quotation marks omitted), or if despite the legal error, “the agency's path may reasonably be discerned,” Treichler, 775 F.3d at 1099 (citation and internal quotation marks omitted). But here, we cannot discern the agency's path because the ALJ made only a general credibility finding without providing any reviewable reasons why she found Brown–Hunter's testimony to be not credible. See id. at 1103. Although the ALJ summarized a significant portion of the administrative record in support of her RFC determination, providing a summary of medical evidence in support of a residual functional capacity finding is not the same as providing clear and convincing reasons for finding the claimant's symptom testimony not credible. We reject the Commissioner's argument, also rejected in Treichler, that because the ALJ “set out his RFC and summarized the evidence supporting his determination” we can infer “that the ALJ rejected [petitioner's] testimony to the extent it conflicted with that medical evidence.” Id. We cannot review whether the ALJ provided specific, clear, and convincing reasons for rejecting Brown–Hunter's pain testimony where, as here, the ALJ never identified which testimony she found not credible, and never explained which evidence contradicted that testimony. Burrell, 775 F.3d at 1138. This “falls short of meeting the ALJ's responsibility to provide ... ‘the reason or reasons upon which’ [her] adverse determination is based.” Treichler, 775 F.3d at 1103, quoting 42 U.S.C. § 405(b)(1).

In sum, “we cannot substitute our conclusions for the ALJ's, or speculate as to the grounds for the ALJ's conclusions. Although the ALJ's analysis need not be extensive, the ALJ must provide some reasoning in order for us to meaningfully determine whether the ALJ's conclusions were supported by substantial evidence.” 775 F.3d at 1103 (internal citation omitted). The ALJ provided no such reasoning here. “Because ‘the agency's path’ cannot ‘reasonably be discerned,’ Alaska Dep't of Envtl. Conserv. [v. EPA], 540 U.S. [461,] 497, 124 S.Ct. 983

[157 L.Ed.2d 967 (2004) ]

, we must reverse the district court's decision to the extent it affirmed the ALJ's credibility determination.” 775 F.3d at 1099.

IV.

10 Brown–Hunter argues that in light of the ALJ's error, we must credit her testimony as true and remand to the district court with instructions to remand to the agency for an immediate award of benefits. A remand for an immediate award of benefits is appropriate, however, only in “rare circumstances.” Id. Before ordering that extreme remedy, we must first satisfy ourselves that three requirements have been met.

First, we must conclude that “the ALJ has failed to provide legally sufficient reasons for rejecting evidence, whether claimant testimony or medical opinion.” Garrison v. Colvin, 759 F.3d 995, 1020 (9th Cir.2014).

Second, we must conclude that “the record has been fully developed and further administrative proceedings would serve no useful purpose.” Id. This requirement will not be satisfied if “the record raises crucial questions as to the extent of [a claimant's] impairment given inconsistencies between his testimony and the medical evidence in the record,” because “[t]hese are exactly the sort of issues that should be remanded to the agency for further proceedings.” Treichler, 775 F.3d at 1105. Importantly, we are “to assess whether there are outstanding issues requiring resolution before considering whether to hold that the claimant's testimony is credible as a matter of law.” Id. This is because “a reviewing court is not required to credit claimants' allegations regarding the extent of their impairments as true merely because the ALJ made a legal error in discrediting their testimony.” Id. at 1106. The touchstone for an award of benefits is the existence of a disability, not the agency's legal error. To condition an award of benefits only on the existence of legal error by the ALJ would in many cases make “ ‘disability benefits ... available for the asking, a result plainly contrary to 42 U.S.C. § 423(d)(5)(A).’ ” Id., quoting Fair v. Bowen, 885 F.2d 597, 603 (9th Cir.1989).

Third, we must conclude that “if the improperly discredited evidence were credited as true, the ALJ would be required to find the claimant disabled on remand.” Garrison, 759 F.3d at 1021.

Finally, even if all three requirements are met, we retain “flexibility” in determining the appropriate remedy. Id. We may remand on an open record for further proceedings “when the record as a whole creates serious doubt as to whether the claimant is, in fact, disabled within the meaning of the Social Security Act.” Id.

Here, although we conclude that the ALJ committed legal error by failing to specify which testimony she found not credible and why, we will not remand for an immediate award of benefits because we are not satisfied that “further administrative proceedings would serve no useful purpose.” Id. at 1020. Indeed, the record raises crucial questions about the extent to which Brown–Hunter's pain and accompanying symptoms render her disabled. Brown–Hunter's testimony that her back and leg pain are relieved only by lying down four to five times a day for thirty minutes each, and by elevating her legs twice a week for thirty minutes to an hour each, appears dispositive to the ultimate disability determination in light of the vocational expert's hearing testimony. Indeed, when asked hypothetically if a person who was required to be “off task approximately 15 to 20 percent of the workday” would be able to sustain the specified light work, the vocational expert testified that he or she “would not.” When asked whether “an individual [who] needed to rest or lie down ... two or three times a day for thirty minutes at a time” on an unpredictable schedule would be able to sustain “any other jobs,” the vocational expert testified, “No.” And even when the ALJ reduced the hypothetical frequency of feet-elevation to only twice a week, the vocational expert again confirmed that such an individual “wouldn't be able to sustain [light work]. No.”

But Brown–Hunter's need to lie down and to elevate her feet with such frequency is contingent on the assumption that her pain is in fact debilitating and that no other alternative treatment besides lying down adequately alleviates that pain. However, “the record raises crucial questions” on these issues “given [the] inconsistencies between [her] testimony and the medical evidence in the record.” Treichler, 775 F.3d at 1105. For example, Dr. Massrour's June 2010 assessment appears to undermine Brown–Hunter's allegation that she had zero residual functional capacity due to debilitating lower back pain. Dr. Massrour found that her pain averaged a 5 on a 10 point scale, that she “denied progressive neurological deficits, focal weakness, [and] dense numbness,” and that she was “independent with [her activities of daily living].” The record also suggests that her pain was adequately controlled with medication. For example, the records of her monthly visits to the Phoenix Pain Management Center indicate that the pain medications she was taking provided “adequate pain control” and that her “[f]unctionability on [t]reatment” was improving with treatment, as evidenced by numbers of 6 to 8 on a 10–point scale, with higher numbers indicating improvement. Another pain management report on May 8, 2009, states that Brown–Hunter reported “adequate” pain control, and that she was “able to function well.”

11 Although medical reports of adequate pain control on medication do not foreclose the possibility that Brown–Hunter still needs to lie down as often and as unpredictably as she alleged, they do create a significant factual conflict in the record that should be resolved through further proceedings on an open record before a proper disability determination can be made by the ALJ in the first instance. See Treichler, 775 F.3d at 1106–07; Burrell, 775 F.3d at 1141–42. “Where there is conflicting evidence, and not all essential factual issues have been resolved, a remand for an award of benefits is inappropriate.” Treichler, 775 F.3d at 1101. We therefore vacate the judgment of the district court and remand with instructions to remand to the agency for further proceedings on an open record.

VACATED AND REMANDED.