You have probably heard that facts backfire, that correcting someone only digs them in deeper. It is a tidy, cynical story, and it is mostly wrong. Minds do change. They just change slowly, unevenly, and least of all on the beliefs tied to who we are.
To judge how well people change their minds, it helps to start with what doing it well would even look like. A belief is a stance toward a claim, held with some level of confidence, and "updating" just means moving that confidence up or down as new evidence comes in. The benchmark psychologists measure people against is the Bayesian one: you should shift your confidence in proportion to how much the new evidence actually bears on the question. Strong evidence should move you a lot, weak evidence barely at all, and what we loosely call "changing your mind" is really that confidence sliding far enough to tip from one side to the other.
Two facts about real minds complicate this. First, beliefs do not sit in isolation; they hang together in coherence networks, where each belief is supported by, and supports, others (Thagard, 1989). Accepting a new claim that clashes with a dozen existing ones is expensive, because it forces a cascade of other revisions, which is part of why big beliefs are sticky and small ones are cheap to change. Second, people are not neutral judges of evidence. Kunda's classic account of motivated reasoning showed that we reason toward conclusions we want to reach, but, crucially, only as far as we can build a plausible justification for them (Kunda, 1990). The wish is not enough; we still have to convince ourselves it is reasonable. That bound matters, and we will come back to it.
At the extreme, disconfirmation can even deepen commitment. The founding example is a 1950s study of a doomsday group whose prophecy failed; rather than abandon the belief, the most invested members proselytised harder (Festinger, Riecken and Schachter, 1956). That image, of evidence making belief stronger, has cast a long shadow over how we talk about changing minds. It also set up the field's most instructive mistake.
For about a decade the dominant story was the backfire effect. In an influential set of experiments, correcting a political misperception sometimes left the targeted group believing the falsehood more firmly than before (Nyhan and Reifler, 2010). The finding spread fast, because it fit the mood of the times and a satisfying cynicism: facts are useless, correction backfires, people are hopeless. There was just one problem. When other researchers tried to reproduce it at scale, it mostly vanished. Across 52 issues and more than ten thousand people, the overwhelming pattern was "steadfast factual adherence", people moved toward the facts, even facts that were inconvenient for their side (Wood and Porter, 2019). Three more large experiments found no backlash even where theory predicted it should be strongest (Guess and Coppock, 2020). Backfire is real but rare, an exception rather than the rule, and this is a tidy case study in how a striking result can outrun its evidence (a theme from L0). The honest headline is the hopeful one: correction usually works, a bit.
So if facts mostly do land, why does belief change feel so hard? Part of the answer is that "motivated reasoning" gets over-diagnosed. Two careful challenges are worth holding. Pennycook and Rand (2019) found that people who fall for partisan fake news are not mainly the most biased but the least reflective, more analytic thinkers spotted falsehoods better regardless of which side the lie favoured, suggesting a lot of bad belief is "lazy, not biased". And Druckman and McGrath (2019) point out that what looks like motivated reasoning is often just people starting from different prior beliefs and trusting different sources, which is rational Bayesian updating, not a defect, and is genuinely hard to tell apart from motivated distortion. The lesson is not that motivated reasoning is a myth, it is real and well evidenced, but that it is over-applied as an all-purpose explanation.
Where belief change really does break down is identity. When a belief becomes a badge of group membership, accuracy stops being the only goal and defending the in-group view takes over. The sharpest demonstration is unsettling: on climate change, people with higher science literacy and numeracy were more polarised, not less, the knowledge was used to build better defences of the position their group already held (Kahan et al., 2012). On these identity-fused topics, more information can widen the gap, which is the opposite of what the "just educate people" instinct predicts.
Two cautions. First, much of this evidence is about immediate change, and immediate updates often fade, so durable belief change is harder than a one-shot experiment suggests. Second, the literature leans heavily on political topics, survey measures, and Western samples, and a correction in a tidy study is not the same as a belief shifting in a noisy life (L0). The reassuring "facts mostly work" finding is real, but it is a floor, not a guarantee.
What makes a correction stick rather than fade? How can we separate genuine motivated distortion from people rationally weighting different priors and sources? And when a belief has fused with identity, what actually loosens it, since more facts often will not?
The usable core: evidence usually moves people a little, so correction is worth attempting, but the beliefs that matter most are often identity-anchored, and those shift through trust, narrative, and framing rather than facts alone.
Sort the beliefs you are trying to change. A peripheral belief ("this software is slow") is cheap to move with a clear demonstration. An identity-anchored one ("people like me do not use brands like that") will not budge on specs, it needs social proof, a trusted messenger, and a story the customer can adopt without feeling they have changed sides. Map which kind you are facing before you write the message. And drop the fear of backfire: correcting a false impression about your product overwhelmingly helps rather than entrenches it, so address misinformation directly rather than staying silent.
Two practical points fall out of the research. Fact-checking is worth doing, it moves more people than the cynics claim, but it does almost nothing on the handful of issues that have fused with partisan identity, where more sophistication just means better counter-arguing. And because repetition itself breeds belief, the cardinal rule of correction is to repeat the true claim, not the myth: restating "the myth that X" to debunk it can leave the myth more familiar, and familiarity feels like truth. Inoculating people before they meet a falsehood works better than chasing it afterwards.
Public communication should be confident that correction is worthwhile, the backfire fear has been overblown, while being realistic that identity-loaded topics (vaccines for some groups, climate, anything tribal) are the hard cases. The tools that work there are not more data but trusted in-group messengers, framing that does not force people to choose between a fact and their identity, and prebunking known misinformation before it spreads. Leading with shared values, then evidence, beats leading with evidence that feels like an attack.
Three habits. First, do not assume facts backfire, they usually help a little, so it is worth correcting the record. Second, check whether the belief you are facing is peripheral or fused to identity, because only the first kind moves on evidence; the second needs trust and a face-saving story. Third, when you correct something, lead with the truth and repeat it, rather than repeating the myth you are trying to kill.
.png)