Why Can Positive Feedback Feel Less Believable During Depression?

Most people can recognize some version of this experience, especially after a stressful event or in an area where they already feel uncertain. During depression, however, negative conclusions about the self can become unusually persistent. A compliment may be heard and understood without producing much movement in the larger belief: “I am not capable,” “I disappoint people,” or “success was an exception.”

It would be inaccurate to turn that observation into a universal rule. People with the same diagnosis differ widely, feedback occurs in different relationships and contexts, and criticism can sometimes be valid while praise can sometimes be vague. The research question is therefore not whether a “depressed brain rejects praise.” It is whether depressive symptom severity is associated with the amount by which new positive or negative information changes an existing self-belief.

A self-belief is not necessarily a fixed declaration such as “I am competent” or “I am unlikeable.” It can also be a quieter expectation: how well you think you will perform, how others are likely to respond, or what a new result says about your ability. These expectations are repeatedly tested against experience.

Suppose you expect to perform at an average level and receive feedback suggesting that you did much better. The difference between the expected outcome and the feedback or outcome that is received is called a prediction error. In ordinary learning, that mismatch provides a reason to update the expectation before the next attempt. If the result was only slightly surprising, the adjustment may be small; if it was credible and very unexpected, the adjustment may be larger.

The 2026 study in Translational Psychiatry included 35 adults with diagnosed depression and 32 matched participants without mental illness. The clinical sample was deliberately recruited with variation in depressive and social-anxiety symptoms. Participants completed a task known as Learning Of Own Performance, or LOOP, while researchers recorded fMRI data.

The task was presented as an experiment in cognitive estimation. Participants estimated quantities such as the weight of an animal, the height of a building, the distance traveled by a vehicle, or an amount of food. Before each answer, they predicted how well they would perform. They then received performance feedback framed as a percentile relative to an alleged reference group.

Crucially, the feedback was controlled by the experiment rather than being a simple report of true ability. In some categories it was predominantly positive and in others predominantly negative. Participants completed trials about their own performance and also observed another person’s supposed performance, allowing the researchers to distinguish learning about the self from learning about someone else.

Across both groups, participants updated new beliefs about their own abilities more after unexpectedly negative than unexpectedly positive feedback. That bias was specific to beliefs about the self and was not observed in the same way when participants learned about the other person. Yet a categorical comparison—diagnosed depression versus healthy control—did not show that the clinical group learned in a fundamentally different way.

The more revealing pattern appeared within the clinical sample. Greater overall symptom burden was associated with a lower learning rate from unexpectedly positive feedback about the self. The correlation was moderate in size (r = −0.51), while updating after unexpectedly negative feedback was not related to symptom burden. In other words, severity tracked weaker positive updating, not stronger negative updating, in the behavioral data.

The symptom measure requires care in interpretation. It combined variation across depressive symptoms, negative automatic thoughts, social anxiety, and self-related measures rather than isolating one pure feature of depression. The result is therefore an association within a small, clinically complex sample—not proof that depressive symptoms alone caused the learning pattern.

If someone already expects failure, ordinary success may not be enough to challenge the expectation. Unexpectedly positive feedback creates a stronger test because it conflicts with the prior belief. In principle, the mismatch supplies exactly the information needed to revise an excessively negative self-view.

The new study suggests that this corrective step may weaken as symptom burden becomes greater. The feedback still arrives, but less of it is carried into the next expectation. A possible psychological explanation is that a firmly established negative model makes contradictory information seem less representative: “They were being kind,” “the task was unusually easy,” or “one good result does not count.” Researchers sometimes describe this protection of an existing belief as cognitive immunization.

That explanation is plausible, but the experiment did not directly demonstrate why each participant updated less. It did not ask whether positive feedback was dismissed as insincere, attributed to luck, forgotten, or outweighed by another concern. Those are candidate mechanisms for future study, not interchangeable facts established by the learning-rate result.

While participants processed feedback, fMRI measured changes in blood oxygenation associated with neural activity. Compared with controls, the clinical group showed a different balance in how the right anterior and posterior insula tracked negative relative to positive prediction errors. Follow-up analyses pointed toward a stronger response to unexpectedly negative feedback, particularly in the posterior insula.

The insula participates in several overlapping functions, including detecting personally relevant events, integrating bodily and emotional signals, and coordinating attention with other large-scale networks. It is therefore reasonable to interpret the result as evidence that negative self-related surprises were processed differently at the neural-systems level. It would not be reasonable to call the insula a “depression center” or to conclude that one scan reveals what a particular person believes.

The neural and behavioral results also should not be collapsed into one causal chain. The study observed both stronger insula sensitivity to negative prediction errors at the group level and reduced positive updating with greater symptom burden inside the clinical sample. It did not establish that the insula response caused the weaker use of positive feedback. Associations between insula activity and symptom burden were in the expected direction but were not statistically significant.

CBT is sometimes caricatured as replacing a negative statement with a positive one. Its more relevant connection here is the examination of stable assumptions, interpretations, and the evidence used to maintain them. A therapist and patient may explore not only “What do you believe about yourself?” but also “What counted as evidence for that belief, and what happens when evidence points in the other direction?”

Behavioral experiments can create experiences that test an expectation, while cognitive work can examine explanations that preserve an old belief after a contradictory outcome. That is closer to belief updating than simply instructing someone to accept compliments. Even so, the 2026 LOOP study did not test CBT, did not measure treatment response, and did not show that deliberately increasing positive learning rates would reduce depression symptoms.

The study makes the familiar experience of “praise that changes nothing” more scientifically precise. It suggests that depression may be related not only to the negative content of self-beliefs, but also to the degree by which unexpectedly positive information revises those beliefs. Just as importantly, it shows that the pattern may be dimensional: severity within a clinical group revealed more than a simple comparison between diagnoses.

That conclusion should remain proportional to the evidence. The experiment generated new beliefs about a narrow kind of performance, the clinical sample was small, and the central severity result was correlational. Real feedback carries information about trust, relationships, history, status, and motive that a laboratory percentile cannot reproduce.