A claim can be withdrawn, labeled false, and replaced with accurate information, yet still shape what people remember and how they explain what happened. Psychologists call this residual effect the continued influence of misinformation. It does not mean that corrections are useless. It means that belief, memory, inference, trust, and behavior are different outcomes, and one message may not move all of them equally.
That distinction matters in medicine and science. A person may correctly say that a rumor about a treatment was disproved while still using the rumor's causal story to judge symptoms; a clinician may remember that an early case report was retracted but retain its association when forming a differential diagnosis. A headline can disappear from view while the explanation it supplied remains available in memory.
The practical lesson is more precise than "debunking fails." Corrections usually help, but a bare denial often leaves an explanatory gap. Better corrections make the verified account easy to understand, identify why the original claim was wrong, replace its causal role, and give the audience a memorable route back to the facts.
What the continued influence effect actually measures#
In a typical experiment, participants first read an account containing a critical piece of information. They may learn, for example, that a warehouse fire involved improperly stored materials. Later they are told that the materials were not present. When asked why the fire spread, some participants still refer to the retracted cause even though they can recall the correction.
The important outcome is not simply whether they can repeat the corrected fact. It is whether the false information continues to guide an inference. That is why a knowledge quiz and a reasoning task can produce different answers. Memory can retain both the original claim and its correction, and retrieval does not always select the right one at the moment it is needed.
Walter and Tukachinsky combined 32 studies with 6,527 participants in a meta-analysis published in 2020. On average, correction did not fully remove the influence of misinformation. The amount of persistence varied. Corrections worked better when they formed a coherent account, aligned with the audience's broader understanding, or came from the source that introduced the error, and they worked less well after repeated misinformation, a delay, or attribution to a credible source.
Those findings describe averages across experimental settings. They do not supply a fixed percentage for every rumor, platform, or person. Topic, message design, prior belief, measurement, and social setting all change the result.
Why deleting a claim leaves a hole in the story#
People use mental models to connect events. If the original claim explains who acted, why an outcome occurred, or what should happen next, removing it can break the story's causal chain, and the mind may reuse the discredited explanation because it remains the most complete account available.
"That is false" removes a proposition but supplies no replacement. A stronger correction says what is known and explains how investigators reached that conclusion. If a laboratory error created the finding, describe the error and the corrected measurement. If a correlation was mistaken for causation, name the confounder or the design limitation. If an image came from another event, identify its actual source and date.
An alternative explanation must be supported, not invented merely to make the message satisfying. When the cause remains unknown, the honest replacement is a bounded account of what has been ruled out, what remains plausible, and what evidence would distinguish the possibilities. Uncertainty can fill a model more responsibly than false precision.
Familiarity can outlast source memory#
Repeated statements become easier to process. That fluency can make them feel more familiar and, under some conditions, more credible. A 2026 systematic review and meta-analysis covering 182 studies found a small illusory truth effect after adjustment for publication bias, with substantial variation across studies and message types.
Meanwhile, memory for where a claim came from can fade faster than memory for the claim itself. You may later remember the sentence without remembering that it appeared under a warning; this source-memory problem helps explain why a correction should not give the myth more visual or verbal prominence than the fact.
The answer is not to avoid identifying the error. A correction must be specific enough for your readers to connect it to the claim they encountered, and the useful balance is fact first, one concise reference to the myth, a clear refutation, and fact again. Repeating the false claim many times for rhetorical effect adds familiarity without adding understanding.
Backfire is possible, but it is not the default#
One widely repeated warning says that correcting a false belief will make people defend it more strongly. Some studies have reported such effects in particular contexts. Later reviews and replications, however, found that reliable backfire is much less common than the warning suggests.
The Nature Medicine review of misinformation interventions concluded that concern about backfire should inform message design but should not prevent correction; a correction that prominently features the accurate account is more likely to reduce misbelief than to strengthen it. Silence also has a cost because an uncontested claim can continue circulating and can appear socially accepted.
This does not justify contemptuous or identity-threatening messaging. A technically correct rebuttal can still damage trust, harden group boundaries, or prompt a reader to reject you. Respect, shared values, transparency about evidence, and freedom to ask questions are part of effective communication, even when the factual correction is not negotiable.
Why science and health claims are especially difficult#
Science is provisional in a disciplined sense: estimates change when better measurements, larger samples, or new outcomes arrive. Bad-faith communicators can recast ordinary updating as inconsistency. A correction may then be interpreted as proof that institutions cannot be trusted rather than as evidence that error detection is working.
Health claims also connect to fear, identity, lived experience, and decisions with immediate consequences. A population estimate may conflict with one vivid story. Mechanistic language can sound more convincing than a randomized comparison even when the mechanism has not predicted patient outcomes. Technical uncertainty can be misread as ignorance, while overconfident reassurance can fail when guidance later changes.
A useful correction therefore distinguishes what changed from what did not. It names the study population, outcome, time horizon, and degree of certainty. It explains whether new evidence reversed a conclusion, narrowed its scope, or only changed the size of an effect, and that structure protects against the false binary that science either knows everything permanently or knows nothing at all.
A fact-first correction in five moves#
Begin with the verified conclusion in plain language. Your readers should be able to understand the correction even if they stop after the first sentence. Add the evidence source and date when recency matters.
Next, signal the false claim once. A warning such as "A circulating post incorrectly states..." helps the reader bind the correction to the right item. Do not turn the myth into the headline if the accurate statement can carry it.
Third, explain the error. Was the number fabricated, the denominator omitted, the study misread, the time order reversed, or the image relabeled? Naming the technique teaches a transferable check rather than asking for trust alone.
Fourth, provide the supported alternative account. The replacement should answer the question the myth appeared to answer. If investigators do not yet know, describe the active uncertainty and the next reliable checkpoint.
Finally, restate the fact and give an action. Link to the primary record, explain how to verify future updates, or state what decision follows now. A correction is more useful when it tells your readers where accurate information lives after the social post scrolls away.
Reach and timing can matter as much as wording#
The original claim and the correction rarely travel through identical networks. A sensational assertion may reach millions before a careful update is issued. People who saw the first message may never see the second. Aggregate belief can therefore remain high even if the correction works well among those who receive it.
Speed helps, but speed cannot justify guessing. Communicators can issue an interim notice that a claim is unverified, preserve the evidence trail, and give a time for the next update. Prebunking, which teaches a misleading technique or gives accurate context before a falsehood arrives, may reduce the amount of later repair required.
Platforms and institutions also shape the outcome. Search results that preserve obsolete pages, screenshots detached from corrections, recommendation systems, repeated broadcast clips, and incentives for outrage can keep an error available. Individual message design cannot compensate for every distribution problem.
How to tell whether a correction worked#
Counting views or positive reactions is insufficient. Ask whether people recognize the false claim, remember the verified account later, use it in reasoning, distinguish accurate from inaccurate new claims, and make better-informed decisions. A message can improve factual recall while leaving behavior unchanged, or reduce belief while increasing generalized skepticism.
Measurement should also look for unequal reach. If your correction only reaches people who already trust the source, a favorable average may conceal failure in the audience most affected. Testing multiple messages, languages, reading levels, and channels can reveal where the account breaks down.
The target is calibrated understanding, not forced agreement. A successful correction enables a reader to state the best-supported conclusion, its limits, and what evidence would warrant another update.
What not to infer from this research#
The continued influence effect does not prove that people are irrational or incapable of learning. Memory is reconstructive, attention is limited, and causal explanations are useful. The same systems that let people reason from incomplete information can preserve an obsolete input.
It also does not show that every disagreement results from misinformation. Values, risk tolerance, institutional history, and reasonable interpretations of uncertain evidence can differ. Labeling all dissent as a cognitive failure is both inaccurate and corrosive.
Finally, the evidence does not support abandoning fact-checking. Corrections generally move beliefs in the right direction. The research asks you to design them as replacements for flawed models, deliver them through trusted routes, and assess what people actually carry forward.
Sources and further reading
- Walter and Tukachinsky meta-analysis of continued influence, Communication Research 2020
- Lewandowsky and colleagues review of misinformation correction
- Nature Medicine review of misinformation susceptibility and interventions
- National Academies report, Understanding and Addressing Misinformation About Science
- 2026 meta-analysis of the illusory truth effect, Nature Communications
- Meta-analysis of corrections for science-relevant misinformation, Nature Human Behaviour 2023
Questions and answers
Does correcting misinformation usually make belief worse?
No. Backfire has occurred in some studies and contexts, but later reviews find it is not the usual response. A clear, respectful, fact-first correction is generally better than leaving a consequential false claim unanswered.
Why can someone remember a correction and still use the false claim?
Remembering that a statement was corrected and retrieving the right explanation during reasoning are different tasks. The original claim may remain linked to a causal story, especially when the correction provides no replacement.
Should a correction repeat the myth at all?
Usually it should identify the myth briefly so readers know what is being corrected. The accurate account should receive greater prominence, and unnecessary repetition of the false wording should be avoided.
What makes an alternative explanation useful?
It fills the same causal role with evidence. It explains what actually happened, why the earlier account was wrong, or what is known and unknown when no single cause has been established.
Is debunking enough to solve misinformation?
No. Corrections are one tool. Early warnings, media literacy, trusted local messengers, transparent institutions, accurate search results, distribution changes, and incentives for reliable communication all matter.