A sentence has been corrected.
The person accepts the correction.
Later, when asked to explain what happened, part of the old story returns.
This small cognitive residue sits near the center of Stephan Lewandowsky's long research programme on misinformation.
The problem is not simply whether people can be told that a claim is false. Often they can. The deeper problem is how information becomes integrated into an explanatory model, what happens when one element of that model is retracted, and why belief updating can be incomplete even when the correction itself is understood.
That distinction has helped move misinformation research beyond a crude picture in which falsehood disappears the moment a fact-check arrives.
Current research identity
As of September 2026, Stephan Lewandowsky is Chair in Cognitive Psychology at the University of Bristol's School of Psychology and Neuroscience.
He also heads Bristol's Technology, Democracy, and Cognition Group, or TeDCog, an interdisciplinary group studying pressure points between human cognition and online technologies.
His current institutional profile describes research on misinformation persistence, conspiracy beliefs, online information architectures and the ways platform systems can shape what people encounter.
That contemporary emphasis grew from a much older cognitive-science question: how does a mind revise what it thinks it knows?
The continued influence effect made correction a memory problem
Lewandowsky's work is strongly associated with the continued influence effect.
The term describes a pattern in which misinformation can continue influencing reasoning after it has been retracted or corrected.
A person may explicitly remember that the original claim was false yet still use elements of it when explaining a later event.
This is not the same as stubbornly insisting that the correction is wrong.
It is a problem of how causal and narrative models are updated.
If misinformation explains why something happened, simply removing the false element can leave a hole. A correction that supplies a coherent alternative may therefore be more effective than one that merely says, "that was wrong."
Correction does not usually make things worse
Public discussion of misinformation has sometimes amplified a fear that corrections routinely "backfire" and strengthen false beliefs.
The broader evidence does not support treating such backfire effects as a general rule.
Lewandowsky and colleagues have been part of the research movement clarifying that corrections are often effective, even if their effects are incomplete or context dependent.
This matters because two different statements are easy to confuse.
One is that misinformation can retain some influence after correction.
The other is that correction increases belief in the misinformation.
The first has substantial empirical support in particular paradigms. The second is not a default consequence of fact-checking.
Memory needs a replacement, not only an eraser
The 2012 review by Lewandowsky, Ullrich Ecker, Colleen Seifert, Norbert Schwarz and John Cook helped systematize psychological research on misinformation and correction.
One practical implication was that effective corrections often do more than negate.
They help the reader build an alternative account.
If a false claim served as a causal bridge in a story, a replacement explanation can make updating easier because the mental model remains coherent.
This does not mean every correction must tell a complete new story.
It means that information is remembered relationally. Removing one proposition can leave the surrounding structure intact.
Source and repetition complicate correction
People do not evaluate claims in a vacuum.
They remember who said something, how often they have encountered it, whether it fits prior knowledge, how fluent it feels and whether trusted others endorse it.
Repeated claims can become easier to process. Familiarity can be mistaken for truth. Source memory can weaken over time even when the content remains available.
Lewandowsky's programme places correction inside this larger architecture rather than treating misinformation as a single defective proposition.
That broader framing matters in contemporary digital environments, where the same claim can appear through many channels while the original source becomes difficult to reconstruct.
The 2022 review widened the lens
A major 2022 Nature Reviews Psychology article led by Ullrich Ecker with Lewandowsky and colleagues synthesised work on the psychological drivers of misinformation belief and resistance to correction.
The review distinguishes multiple stages and influences: exposure, source characteristics, prior beliefs, affect, social identity, memory processes, correction design and information environments.
It also reviews both proactive strategies such as prebunking and reactive strategies such as debunking.
The important contribution is not a single recipe.
It is the rejection of one-cause explanations.
Misinformation is not explained adequately by saying that people are irrational, partisan, inattentive or gullible. Different mechanisms can dominate in different contexts.
The continued influence effect is not one mechanism
Even within correction research, several mechanisms have been proposed.
One family of explanations emphasizes mental-model updating: misinformation has already been integrated into a coherent representation, and correction leaves a gap unless an alternative is supplied.
Another emphasizes retrieval competition: both the misinformation and correction may remain in memory, and later reasoning depends on which representation is retrieved.
Source-monitoring failures, familiarity and motivational processes can also contribute.
A careful profile should therefore resist turning the continued influence effect into evidence for one settled cognitive mechanism.
The effect is an empirical pattern. Its causes can vary.
The Debunking Handbook translated evidence into guidance
Lewandowsky was among the authors of The Debunking Handbook 2020, a consensus-oriented practical guide produced with researchers in misinformation and correction.
Its recommendations include making truth salient, warning when useful, explaining the fallacy or problem, providing a clear correction, and avoiding unnecessary repetition of the false claim.
Such guidance is valuable because practitioners need usable principles.
But a handbook is not a universal protocol.
Effects depend on topic, audience, medium, prior belief, social context and the quality of the correction.
A research-grounded Institute should preserve the distinction between evidence-informed guidance and guaranteed intervention.
Lewandowsky's programme increasingly includes information architecture
His current Bristol work extends beyond individual memory.
TeDCog studies the interaction between cognition and online technology, including how digital platforms, recommender systems and information environments can affect democratic discourse and autonomy.
This expansion matters conceptually.
A misinformation problem can be framed at several levels:
- the proposition is false;
- the person's memory is imperfect;
- the source ecosystem is polluted;
- the interface repeatedly privileges some claims;
- incentives reward emotionally engaging content;
- corrective information arrives too late or too weakly.
The more the problem is distributed across levels, the less plausible it becomes to solve it through individual media literacy alone.
Human correction lives inside institutional systems
Imagine a public-health agency correcting a false claim.
Its factual wording can be excellent and still fail if the audience distrusts the institution, if the correction travels slowly, if headlines repeat the false claim without context, or if people have already built social identities around the disputed issue.
Lewandowsky's research helps connect the cognitive mechanics of belief updating with the institutional conditions in which correction occurs.
For the Institute, that is an important bridge from individual discernment to public information infrastructure.
Inner Technology cannot be reduced to teaching a person to "spot misinformation" while leaving the surrounding systems untouched.
Correction is also an epistemic design problem
A good correction has at least two jobs.
It must improve the factual model.
It must also make future reasoning less dependent on the false information.
Those goals are related but not identical.
Someone can answer a direct belief question correctly and still use misinformation indirectly in inference.
Someone else can update their causal model without remembering the precise wording of the correction.
Measurement therefore matters.
Belief ratings, memory tests, inferential questions, sharing intentions and behavior are different outcomes. A study that improves one should not be described automatically as improving all of them.
The European research context is part of the programme
Lewandowsky's present work is embedded in European research and policy collaborations on online information, autonomy and democratic resilience.
Bristol's TeDCog group explicitly studies the relationship between cognition, technology and democratic systems, and Lewandowsky has participated in European-level policy work for years.
This is relevant to the Institute's Northern European position, but it should not be used as prestige decoration.
The substantive connection is that Europe increasingly treats information integrity, platform governance, media literacy and democratic resilience as linked systems problems.
Lewandowsky's cognitive research gives that policy conversation a mechanism-level foundation.
Synthetic media raises a new correction problem
Generative AI makes false claims easier to produce at scale and can also generate persuasive explanations around them.
But the important cognitive problem may not be sheer volume alone.
Synthetic content can create ambiguity about provenance, source identity and whether a correction addresses the same object the user originally encountered.
A person may see a fabricated image, an AI summary of commentary about the image, a correction of the summary and a repost stripped of the correction.
The resulting memory trace is no longer one proposition plus one retraction.
It is an information chain.
Lewandowsky's work does not yet settle how correction functions in every generative-AI environment. It provides a conceptual toolkit for asking the question precisely.
Timing changes what a correction has to compete with
A correction delivered seconds after exposure enters a different memory environment from one delivered days later.
Early correction may prevent a false causal model from stabilising. Later correction may need to compete with repetition, discussion and secondary sources that have already reinforced the claim.
This means "correction effectiveness" is incomplete without timing.
Researchers should report the interval between misinformation and correction, the interval between correction and outcome measurement, and whether participants encountered the claim repeatedly in between.
The same correction can look strong in an immediate post-test and weak in a delayed reasoning task without either result being contradictory.
Repetition creates a communication dilemma
To correct a false claim, communicators often need to identify it.
But repeating misinformation can increase familiarity, and familiarity can influence later judgments.
This has encouraged practical formats that foreground accurate information, mention the false claim only as needed, explain why it is wrong and return to the accurate account.
The principle is not a magical word order.
It is an attempt to ensure that the correction leaves a stronger and more usable representation than the misinformation itself.
Practitioners should therefore avoid turning "truth sandwich" or similar labels into rigid formulas detached from audience and context.
Distribution can overwhelm a cognitively effective correction
Suppose a correction reduces belief among 60% of the people who actually see it.
If the misinformation reaches ten million people and the correction reaches fifty thousand, cognitive efficacy and population impact diverge.
This is one reason Lewandowsky's movement toward information architecture matters.
The quality of correction is only one component. Reach, timing, amplification and repeated exposure also determine what an information ecosystem remembers.
Institutions should therefore evaluate both per-person effect and distributional coverage.
A perfect correction nobody encounters is not an effective public information system.
Institutional memory needs correction trails
Organisations also suffer from continued influence.
A report is circulated, a figure is later revised, but old slides remain in shared drives and downstream documents keep citing the earlier number.
Correction systems can therefore include provenance practices: version labels, visible change logs, linked retractions, machine-readable update notices and active notification of teams that used the prior information.
These are not findings from Lewandowsky's laboratory paradigms directly.
They are an Institute synthesis inspired by the same problem: a correction must reach the places where the old representation continues to operate.
Correction quality should be measured at more than one time point
Immediate belief change can overstate durable success.
A correction may be memorable in the minutes after exposure because it is the most recent information in working memory. The more demanding test comes later, when the person must retrieve the update without the correction sitting in front of them.
Studies can therefore separate immediate belief ratings, delayed belief ratings and later inferential reasoning. These outcomes may diverge.
For institutional practice, the same logic suggests re-surfacing corrections when old claims predictably reappear rather than assuming one notice permanently repairs the information environment.
Correction can threaten identity without being factually weak
A correction sometimes enters a social conflict rather than a neutral memory task.
The recipient may experience the claim as connected to group identity, loyalty or moral standing. In those contexts, the factual content of the correction and the social meaning of accepting it can pull in different directions.
Lewandowsky's broader synthesis work includes motivational and social influences precisely because memory updating does not happen outside relationships.
This does not justify accepting falsehood for the sake of belonging. It explains why effective correction may require credible messengers, respect and opportunities to revise without public humiliation.
Those are contextual hypotheses, not substitutes for factual accuracy.
What the evidence does not show
Research on the continued influence effect does not show that misinformation is impossible to correct.
It does not show that every false claim leaves a durable cognitive residue.
Correction studies do not establish that one format works equally well across cultures, topics or populations.
The existence of a continued influence effect in an experiment does not by itself show that the same magnitude will occur in a live information ecosystem.
Evidence that online architectures shape exposure does not mean platform design explains every false belief.
And Lewandowsky's prominence in misinformation research is not itself evidence for the validity of a particular mechanism or policy recommendation.
What this changes for Inner Technology
Lewandowsky's work makes discernment less romantic.
A person can care about truth, accept a correction and still reason with residue from the earlier story.
The relevant capacity is therefore not simply skepticism.
It includes the ability to update an explanatory model, retrieve the correction when it matters, tolerate the discomfort of revision and remain attentive to source and context.
Institutions have parallel responsibilities.
They need correction systems that are timely, coherent, legible and designed for memory rather than merely publication.
The sentence in the margin
Return to the corrected page.
The old claim is crossed out.
The replacement is accurate.
But the mind that read the first version is not a blank page either.
It contains traces, expectations and causal structure.
Lewandowsky's research programme asks what it takes for a correction to reach that deeper level of revision.
Not simply to replace a sentence.
To change what the person will use when they next have to think.
Research references
- Lewandowsky S, Ecker UKH, Seifert CM, Schwarz N, Cook J. Misinformation and its correction: continued influence and successful debiasing. Psychological Science in the Public Interest. 2012;13(3):106-131. https://doi.org/10.1177/1529100612451018
- Ecker UKH, Lewandowsky S, Cook J, et al. The psychological drivers of misinformation belief and its resistance to correction. Nature Reviews Psychology. 2022;1:13-29. https://doi.org/10.1038/s44159-021-00006-y
- Lewandowsky S, Cook J, Ecker UKH, et al. The Debunking Handbook 2020. George Mason University Center for Climate Change Communication; 2020. https://sks.to/db2020
- Lewandowsky S, Ecker UKH, Cook J. Beyond misinformation: understanding and coping with the post-truth era. Journal of Applied Research in Memory and Cognition. 2017;6(4):353-369. https://doi.org/10.1016/j.jarmac.2017.07.008
- University of Bristol. Professor Stephan Lewandowsky institutional profile. Current role verified September 2026. https://www.bristol.ac.uk/people/person/Stephan-Lewandowsky-ebfa6836-6895-4acf-b57b-b2acdf30ec99/
- University of Bristol. Technology, Democracy, and Cognition Group (TeDCog). Current group description verified September 2026. https://www.bristol.ac.uk/psychology-neuroscience/research/research-themes/self-and-society/tedcog/