A person watches a ninety-second video about manipulation.
The video explains one tactic: a false dilemma, perhaps, or emotional language designed to overwhelm judgment.
Hours later, the person sees a new message using the same kind of move.
Can they recognize it?
Do they become more accurate, or merely more suspicious?
Can the effect survive outside the laboratory?
These questions sit at the center of Sander van der Linden's research programme.
Current research identity
As of September 2026, Sander van der Linden is Professor of Social Psychology in Society at the University of Cambridge and Director of the Cambridge Social Decision-Making Lab in the Department of Psychology.
His programme spans misinformation, social influence, judgment and decision-making, with psychological inoculation as one of its most visible lines of work.
Current Cambridge materials also show the programme extending toward AI-generated persuasion, coordinated inauthentic activity and the interaction between misinformation research and new generative systems.
The page is therefore not a biography of a public expert.
It is a map of a research programme.
From persuasion theory to misinformation intervention
Van der Linden's inoculation work draws on a much older persuasion tradition associated with William McGuire.
The contemporary move was to ask whether people could be prepared for misinformation by exposing them to weakened examples of misleading techniques before they encountered stronger versions.
This became widely known as prebunking.
The key innovation in van der Linden's programme was to turn a theoretical idea into scalable digital interventions: online games, short videos and cross-cultural tests.
That practical translation is a major reason the research became influential.
The Bad News game made the theory interactive
One early intervention line involved the game Bad News, developed with collaborators including Jon Roozenbeek.
Instead of only telling users that misinformation exists, the game lets players enact common manipulation tactics in a simulated social-media environment.
The design logic is experiential: by seeing how impersonation, emotional manipulation, polarization or conspiracy framing can be constructed, people may become better at recognizing those techniques later.
That is a distinctive contribution.
It also raises a central methods question: what exactly is learned through the game?
A tactic category? A skeptical stance? A counterargument? A memory for examples?
Cross-cultural testing matters because manipulation is not culturally neutral
A 2020 study tested prebunking interventions based on inoculation theory across several cultural and linguistic settings.
The finding that effects could appear across countries helped move the work beyond one English-speaking convenience sample.
But cross-cultural replication should not be confused with cultural universality.
Examples, humor, trust in institutions, rhetorical conventions and political histories differ.
Translation can preserve words while changing social meaning.
The research programme is strongest when it treats cultural transfer as an empirical question rather than a scaling assumption.
The 2022 Science Advances work tested scale directly
A major step came in 2022, when Roozenbeek, van der Linden and colleagues reported seven preregistered studies, including six randomized controlled studies and a field study on YouTube.
The interventions used short videos teaching manipulation techniques such as emotionally manipulative language, incoherence, false dichotomies, scapegoating and ad hominem attacks.
The studies reported improvements in outcomes including technique recognition, confidence in spotting techniques, discernment and sharing quality under the tested conditions.
The YouTube field component mattered because it moved exposure into a noisy real platform environment.
Scalable does not mean complete
Digital interventions are attractive because a short video can be delivered to millions of people.
Scale changes the public-health arithmetic of an intervention.
A small average effect can matter if reach is enormous.
But scale also introduces heterogeneity.
Some people watch attentively. Others scroll past. Some have seen similar content before. Some encounter the misinformation first. Others receive repeated warnings.
A laboratory effect and a population deployment effect therefore belong to different levels of inference.
Van der Linden's programme increasingly sits at that boundary between experimental efficacy and information-system implementation.
Signal detection helps answer the distrust question
One criticism of misinformation interventions is that they may simply make people distrust all content.
That would look useful if a study only measured lower belief in false claims.
It would not represent improved discernment.
The 2026 signal-detection-theory meta-analysis by Simchon, Zipori, Teitelbaum, Lewandowsky and van der Linden addressed this by separating discrimination ability from response bias across 33 inoculation experiments with 37,075 participants.
The analysis found improved discrimination without a systematic shift toward generalized skepticism.
For the Institute, this is a particularly important contribution because it measures the right failure mode.
Intervention success does not settle the theoretical mechanism
The strongest current critique of the field comes from the 2026 systematic review by Loughnan and colleagues.
Reviewing 72 digital misinformation-inoculation studies, the authors argued that many studies did not adequately test the core theoretical mechanisms of inoculation, especially threat and counterarguing.
That creates a productive tension.
Van der Linden's intervention programme has substantial outcome evidence.
The theory-mechanism link is less fully demonstrated.
A mature profile should hold both facts at once.
The metaphor of immunity is rhetorically powerful and scientifically risky
Van der Linden has often used the metaphor of psychological inoculation or vaccination because it makes preventive intervention intuitive.
The metaphor helped communicate the research to large audiences.
But psychological effects are not biological immunity.
They can decay, fail to transfer or depend on context.
A person who recognizes one tactic is not protected from every false claim.
The Institute therefore preserves the metaphor as historical and explanatory language while refusing to let it carry evidentiary conclusions about permanence.
Technique-based interventions target structure rather than topic
One of the programme's most useful ideas is the emphasis on manipulation techniques rather than only individual false claims.
If the learner can recognize impersonation or false dichotomies across topics, intervention effects may generalize better than claim-specific correction.
That is the promise.
The hard part is measuring genuine transfer.
A new headline that uses the same obvious cue is near transfer.
A subtle, adaptive, contextually different manipulation is a more demanding test.
The research frontier is therefore not simply "does prebunking work?" but what level of abstraction transfers?
Van der Linden's programme is increasingly an AI programme too
Current Cambridge materials describe research on biases in large language models, AI-generated propaganda, personalized AI and system-level coordinated inauthentic activity.
This is a logical extension of the misinformation work.
Generative systems can create persuasive content faster, adapt it to users and vary tactics in response to feedback.
That makes static intervention libraries less obviously sufficient.
The research question becomes whether people can learn principles that remain useful against adaptive machine-generated persuasion.
This is emerging territory and should be labeled as such.
Personalization creates both opportunity and governance risk
An AI system could tailor a prebunking example to a person's language, profession or likely information environment.
That could make training more relevant.
It could also infer psychological vulnerabilities and target them.
The same capability that personalizes resistance training can personalize persuasion.
The Institute therefore treats adaptive inoculation as a dual-use design question.
Effectiveness cannot be separated from consent, transparency and data governance.
Public impact should not be confused with scientific certainty
Van der Linden's tools have been used by governments, international organizations and technology companies.
That level of adoption is important for understanding the programme's reach.
It does not increase the internal validity of a study.
Public uptake, awards and media visibility are profile facts, not evidence grades.
The Institute's profile architecture makes this distinction explicit so that influence is neither ignored nor mistaken for proof.
The programme's strongest methodological habit is comparative measurement
The inoculation work becomes most informative when it includes both manipulative and non-manipulative materials.
That allows researchers to ask whether an intervention improves sensitivity rather than shifting people toward blanket disbelief.
This is a broader lesson for misinformation research.
Interventions should be evaluated on the true information they preserve, not only the false information they suppress.
Discernment is fundamentally a two-sided measurement problem.
Confidence is useful only when paired with accuracy
Some inoculation studies report increased confidence in recognizing manipulation.
That can be beneficial if confidence becomes better aligned with actual performance.
Confidence alone is ambiguous.
An intervention that makes people more certain without making them more accurate could worsen epistemic calibration.
This links van der Linden's programme to the Brier-score and calibration page later in this batch.
The Institute's evidence standard therefore asks whether confidence and accuracy move together.
Institutional deployment changes the ethics
A university teaching manipulation techniques is not identical to a platform deciding what users should be warned about.
At scale, someone selects the tactics, examples and target populations.
That selection has governance implications.
Transparent criteria, symmetric examples and independent evaluation reduce the risk that preventive education becomes a tool for pre-framing contested claims.
The research programme contributes useful intervention science.
Institutions remain responsible for how they operationalize it.
An intervention has more than one outcome layer
A misinformation intervention can succeed at one layer and fail at another.
Participants may recognise manipulation techniques more accurately without changing sharing behavior. They may improve truth discernment while becoming slightly more cautious overall. They may remember the lesson but fail to transfer it to a new format. They may show an immediate effect that disappears weeks later.
For this reason, the strongest evaluation designs separate:
- knowledge of manipulation techniques;
- perceived reliability or accuracy;
- discrimination between higher- and lower-quality content;
- response bias or generalized skepticism;
- sharing intention or behavior;
- delayed retention;
- transfer to unseen examples.
Van der Linden's programme has increasingly used designs capable of distinguishing some of these outcomes. The 2026 signal-detection meta-analysis is especially relevant because it tests whether gains reflect improved discrimination rather than a blanket tendency to distrust.
Population heterogeneity is not noise to be averaged away
Digital interventions are often attractive because they can be deployed at scale.
Scale increases the importance of heterogeneity.
Language, political context, media system, age, educational background, prior knowledge and baseline trust can change how an intervention is received. A manipulation technique that is familiar in one media environment may be obscure in another. Humor may travel differently across cultures. A game framed as playful inoculation may feel patronising to some users.
Cross-cultural validation is therefore not a decorative replication step.
It asks whether the intervention is teaching a portable recognition skill or relying on culturally specific cues.
Deployment changes the causal object
A laboratory or online experiment controls exposure tightly.
A public deployment does not.
People can skip material, encounter competing messages, discuss it socially, forget it, or receive it at a moment when the topic feels personally threatening. Platforms can choose who sees the intervention and when.
Once an intervention enters a live system, effectiveness becomes a joint product of content, delivery, attention, timing and surrounding information.
This creates a useful distinction between intervention efficacy under study conditions and intervention effectiveness in operational environments.
The Institute should keep those terms separate whenever possible.
The public-health metaphor should not hide agency
Inoculation language can make misinformation sound like a pathogen and users sound like hosts.
The metaphor is useful for describing pre-exposure resistance. It is weaker as a model of human agency.
People interpret claims, choose communities, distrust institutions for reasons, produce counterarguments and sometimes strategically share information they do not believe.
Van der Linden's best work is richer than a passive-infection metaphor because it measures discernment and interaction.
The Institute should preserve that richness rather than allowing the metaphor to become the theory.
Long-term use requires a maintenance model
If an intervention is deployed in schools, platforms or public communication, decision-makers need more than an immediate effect size.
They need to know how quickly effects decay, whether brief boosters restore performance, whether repeated exposure creates fatigue and whether users can still apply the technique months later in a different content format.
This suggests a maintenance model similar to other learning interventions: initial acquisition, retention, retrieval, transfer and renewal.
The language of inoculation can obscure this ordinary learning architecture because it implies a one-time protective event. Empirically, repeated practice may turn out to matter more than the metaphor suggests.
Negative side effects should be measured, not assumed absent
A defensible deployment should also look for costs.
Does the intervention reduce trust in accurate content? Does it increase cynicism toward institutions? Does a playful misinformation game trivialise high-stakes harms for some users? Do participants become confident in spotting manipulation without actually improving discrimination?
The 2026 signal-detection evidence helps address generalized skepticism at the aggregate level, but no single meta-analysis can eliminate all context-specific harms.
Prebunking should therefore be evaluated like any other behavioral intervention: benefits, side effects, heterogeneity and opportunity costs all belong in the evidence frame.
What the evidence does not show
Van der Linden's research does not show that people can be permanently immunized against misinformation.
It does not show that every intervention described as inoculation works through the same mechanism.
It does not show equal effectiveness across all cultures, topics or demographic groups.
Short-term gains do not establish durable real-world behavior change.
Improved manipulation recognition does not guarantee correct factual verification.
And the public adoption of a tool does not establish that every claim made about that tool is scientifically settled.
What this changes for Inner Technology
Van der Linden's programme contributes a design principle:
prepare discernment before the difficult message arrives.
That principle becomes useful when paired with measurement discipline.
Did the learner improve discrimination?
Did false positives increase?
Did confidence become better calibrated?
Did the skill transfer to a novel tactic?
Did the effect survive after delay?
Can the learner apply the same standard to messages from sources they already favor?
These questions turn prebunking from a slogan into a trainable, testable capacity.
The second screen
The later message appears.
The participant recognizes something.
Maybe it is the false dilemma. Maybe the emotional manipulation. Maybe only the feeling that the message is pushing too hard.
That recognition is not the end of discernment.
It is the beginning of a verification decision.
Van der Linden's research matters because it has made that tiny interval experimentally visible and technologically scalable.
Its next challenge is to understand with equal precision what happens inside it.
Research references
- van der Linden S, Leiserowitz A, Rosenthal S, Maibach E. Inoculating the public against misinformation about climate change. Global Challenges. 2017;1(2):1600008. https://doi.org/10.1002/gch2.201600008
- Roozenbeek J, van der Linden S. Fake news game confers psychological resistance against online misinformation. Palgrave Communications. 2019;5:65. https://doi.org/10.1057/s41599-019-0279-9
- Roozenbeek J, van der Linden S, Nygren T. Prebunking interventions based on "inoculation" theory can reduce susceptibility to misinformation across cultures. HKS Misinformation Review. 2020;1(2). https://doi.org/10.37016/mr-2020-008
- Roozenbeek J, van der Linden S, Goldberg B, Rathje S, Lewandowsky S. Psychological inoculation improves resilience against misinformation on social media. Science Advances. 2022;8(34):eabo6254. https://doi.org/10.1126/sciadv.abo6254
- Simchon A, Zipori T, Teitelbaum L, Lewandowsky S, van der Linden S. A signal detection theory meta-analysis of psychological inoculation against misinformation. Current Opinion in Psychology. 2026;67:102194. https://doi.org/10.1016/j.copsyc.2025.102194
- Loughnan D, van Stekelenburg A, Pouwels JL, Fransen ML, Kleemans M. An analysis of studies testing digital interventions to inoculate against misinformation: A systematic review. Communication Research. 2026;53(7):1020-1053. https://doi.org/10.1177/00936502251411467
- University of Cambridge, Department of Psychology. Sander van der Linden profile and Social Decision-Making Lab biography. Current affiliation verified September 2026. https://www.psychol.cam.ac.uk/people/sander-van-der-linden