Inoculation theory and prebunking: what the evidence shows

In one line: prebunking warns people in advance about a manipulation technique and shows them a weakened example, on the theory that this builds resistance before the real thing arrives; it reliably teaches people to name techniques in tests, but whether it improves their ability to tell true from false news, and for how long, is disputed in the published record.

Evidence: mixed. Technique recognition is widely replicated, but the claim of better true/false discernment is challenged by signal-detection reanalyses, effects decay without boosters, and one realistic-feed study found no transfer.

Background

William J. McGuire and his Yale collaborator Demetrios Papageorgis set out the idea in experiments published between 1961 and 1964, including papers in Sociometry and the Journal of Abnormal and Social Psychology (1961) and McGuire's 1964 "Inducing resistance to persuasion: Some contemporary approaches". The analogy was medical: a belief that has never met a counter-argument has no defences, so a weakened attack plus its refutation was expected to make it harder to overturn later.

From 2018 researchers at the University of Cambridge Social Decision-Making Lab, led by Sander van der Linden and Jon Roozenbeek, applied the idea to online misinformation. They moved from inoculating against one claim to inoculating against the techniques that recur across many claims, and call this "prebunking". A 2022 review by Cecilie Traberg, Roozenbeek and van der Linden summarises the programme; van der Linden's 2023 book Foolproof presents it to a general audience.

Sander van der Linden speaking at the Hay Festival, 2019

Sander van der Linden, University of Cambridge, at the Hay Festival, 25 May 2019. Photo: LiteraryCritic1, CC BY-SA 4.0, via Wikimedia Commons.

What was tried

Bad News (February 2018). A browser game built by Cambridge researchers with the Dutch media collective DROG and design agency Gusmanson. Players act as a fake-news producer and practise six techniques: polarisation, invoking emotions, spreading conspiracy theories, trolling, deflecting blame and impersonating fake accounts.
Go Viral! (October 2020). A shorter game on COVID-19 misinformation, made by the Cambridge lab with the UK Cabinet Office and DROG. Players were linked to the World Health Organization's COVID "mythbusters" pages, and the WHO and United Nations supported translation beyond the first German and French versions.
Harmony Square (November 2020). A game about political misinformation, made by Cambridge psychologists with support from the US State Department's The State Department's Global Engagement Center (2016 to 2025) and the Department of Homeland Security's Cybersecurity and Infrastructure Security Agency (CISA).
YouTube videos with Google's Jigsaw (published August 2022). Cambridge and University of Bristol psychologists made five 90-second videos on emotionally manipulative language, incoherence, false dichotomies, Scapegoating and ad hominem attacks.
Central and Eastern Europe (from September 2022). Jigsaw ran videos in Poland, Czechia and Slovakia on two narratives about Ukrainian refugees that the project classed as manipulative: one blaming them for the rising cost of living (scapegoating, in Jigsaw's terms; see Scapegoating), and one portraying them as violent (fearmongering, in Jigsaw's terms; see Fear appeal). Narratives were chosen after interviews with local experts, including Demagog and Poland's NASK institute. The videos ran on YouTube, Facebook, Twitter and TikTok.
Germany (announced February 2023, running from 5 June 2023). Jigsaw and the agency Moonshot, with Correctiv, the Alfred Landecker Stiftung, the Amadeu Antonio Stiftung, Das NETTZ, Neue Deutsche Medienmacher:innen and klicksafe, ran videos on decontextualisation, scaremongering and Whataboutism on YouTube, Instagram and Facebook under the line "Don't let yourself be manipulated".
EU elections (2024). Jigsaw's campaign before the 2024 European Parliament elections reached more than 120 million YouTube users, with videos on scapegoating, decontextualisation and discrediting.

The videos named techniques rather than claims. Three of those techniques (whataboutism, fear appeals and scapegoating) also appear in historical propaganda from several countries. The gallery objects below are historical examples from the German, British and Weimar-era record, chosen because they show these three techniques on the page; the gallery holds examples from other countries and eras too. They are not prebunking material; each caption describes only what the object does. See Whataboutism, Fear appeal and Scapegoating.

Imperial German 1916 chart comparing Germany, England and France, issued to answer Allied claims of German barbarism

Whataboutism: "Sind wir die Barbaren?" (Are We the Barbarians?), Imperial German government chart by Louis Oppenheim, 1916. It answers an accusation with a different comparison, of social spending and literacy against Britain's and France's. Sind wir die Barbaren? (Are We the Barbarians?)

British 1915 recruiting poster of a soldier standing between the viewer and a woman and child fleeing a burning village

Fear appeal: "Remember Belgium, Enlist To-Day," UK Parliamentary Recruiting Committee, 1915. The atrocity it invokes is documented; the poster reduces it to one silhouette and a call to enlist. Remember Belgium — Enlist To-Day

Weimar-era 1919 poster: one man strangles another while a woman and infant lie in blood, under the title Spartacus at Work

Scapegoating: "Spartakus bei der Arbeit" (Spartacus at Work), Association to Combat Bolshevism, Berlin, 1919. One political faction is made the author of a generic family murder. Spartakus bei der Arbeit (Spartacus at Work)

Measured effects

Study	Design	N	Finding	Caveats
Roozenbeek and van der Linden 2019, Palgrave Communications	Pre-post ratings around one play of Bad News	about 15,000	Perceived reliability of fake headlines fell by an average of 21% after play (Cambridge release)	Pre-post design; see the response-bias critiques below
Roozenbeek and van der Linden 2020, HKS Misinformation Review	Harmony Square vs control	not recorded here	Players shared fake news at a rate 11% lower than control and, in ScienceDaily's summary, "called bullshit on misinformation 16 percent more often"	Evaluated by the game's developers
Roozenbeek et al. 2022, Science Advances	Six preregistered RCTs plus a YouTube field study	6,464 (RCTs); 22,632 (field)	Videos improved technique recognition, confidence, discernment of trustworthy from untrustworthy content and sharing decisions	Field outcome was one quiz question answered within 24 hours (median 18.4 hours) via YouTube's "brand lift" survey tool
Maertens et al. 2021, J. Exp. Psych.: Applied	Three longitudinal experiments	not recorded here	Significant decay over two months; effect held for at least three months with regular "booster" testing	Boosters were repeated testing, not new content
Maertens et al. 2025, Nature Communications	Five preregistered longitudinal experiments	11,759	Text and video effects lasted about a month; game effects decayed faster; memory, more than motivation, predicted how long effects lasted	Boosters targeted memory of the intervention
Jigsaw 2022 CEE campaign	Ad campaign plus post-exposure survey	over 38 million views, almost a third of the three populations	Survey after two weeks found improved recognition of the tactics among viewers	Company-reported; not a peer-reviewed paper
Jigsaw Germany 2023	Ad campaign plus survey	not published	Share of viewers correctly identifying the tactics rose by up to eight percentage points	Company and partner reporting
12-country EU study 2025, Communications Psychology	13 surveys in 12 EU countries	19,735	Small but significant gains (d ≈ 0.08 to 0.38) in manipulation and technique discernment and sharing decisions; 50-second videos more consistent than 20-second ones	Effects varied with country education index and personal education

Null results, decay and critiques

Response bias versus discernment. Ariana Modirrousta-Galian and Philip Higham (2023, Journal of Experimental Psychology: General) reanalysed five earlier studies with receiver operating characteristic (ROC) analysis from signal detection theory. With comparable true and fake items, Bad News and Go Viral! did not improve discrimination; players answered "false" more often to everything. In 2024 Higham, Modirrousta-Galian and Tina Seabrooke argued that the usual mean-difference measure confounds the two.
Replications with more true items. In "Mixed News about the Bad News Game" (2023, Journal of Cognition), four student teams replicated the drop in belief in false tweets, but belief in true tweets fell by the same or almost the same amount; their exploratory signal detection analysis found increased response bias and no improved discrimination. Seabrooke, Modirrousta-Galian and Higham (2025, Psychonomic Bulletin & Review) tested Bad News with 150 Indian participants in a preregistered study and found no significant change in discrimination or response bias once both counterbalancing conditions were included.
The reply. A signal detection meta-analysis by Simchon, Zipori, Teitelbaum, Lewandowsky and van der Linden (Current Opinion in Psychology, 2026) pooled 33 inoculation experiments with about 37,000 participants in a hierarchical Bayesian model. It reports that gamified and video interventions improved discrimination without increasing response bias. Two of its authors are among the developers of the approach.
Other meta-analytic disputes. A 2023 meta-analysis by Lu and colleagues in the Journal of Medical Internet Research drew a 2025 letter to the editor stating that six sets of extracted data did not represent the effect of interest; the authors published a correction clarifying their methods.
Field-to-feed transfer. Sze Yuh Nina Wang, Samantha Phillips, Kathleen Carley, Hause Lin and Gordon Pennycook (2025, PNAS Nexus) ran five preregistered studies in a simulated social feed. Inoculation cut engagement with emotionally manipulative posts only when the feed held nothing but synthetic posts about that technique; adding real tweets, or posts using other techniques, appeared to remove the effect.
Decay. Go Viral!'s launch publicity said one play of a similar game reduced susceptibility "for at least three months". Maertens et al. (2021) found that duration only with regular booster tests, and significant decay over two months without them.
Data errors. Roozenbeek and colleagues published a 2023 correction to their 2022 Royal Society Open Science paper after pre and post columns for one item were found reversed; means and test statistics changed, and the authors state the conclusions did not.

Who uses it and who criticises it

Users. The UK Cabinet Office (Go Viral!), the WHO (linking Go Viral! to its mythbusters), the US State Department's Global Engagement Center and CISA (Harmony Square), Google's Jigsaw with European NGOs and fact-checkers, and the NATO Strategic Communications Centre of Excellence, which published a report titled Inoculation Theory and Misinformation. Jigsaw's head of research and development, Beth Goldberg, told the Associated Press in 2023: "There's a real appetite for solutions."

Critics on method. Modirrousta-Galian, Higham and Seabrooke (above) argue the measured gains are mainly generic scepticism. Alberto Acerbi, Sacha Altay and colleagues argue that because misinformation is a small share of most people's news diet, raising trust in reliable sources should take priority over fostering distrust in unreliable ones. The philosopher Daniel Williams, reviewing Foolproof, argued that misinformation is not a virus people can be vaccinated against and that, on the book's own narrow definition, misinformation is not widespread. The philosophers Andy Norman and Lee McIntyre replied that Williams' review was "specious" and "one-sided".

Critics on who runs it. The approach's own developers list open ethical questions about who delivers the inoculation (Traberg, Roozenbeek and van der Linden, 2022). Free-expression advocates have mostly treated prebunking as the less restrictive option: The Future of Free Speech, a free-expression think tank, commissioned a 2023 background report from the Aarhus University political scientist Michael Bang Petersen that recommends investing in prebunking. No civil-liberties group or left-wing critic objecting to who runs prebunking campaigns was found on the record for this page. Catherine Salgado of the conservative Media Research Center's NewsBusters described the German campaign in February 2023 as Google's "'Prebunking' Censorship". The Epoch Times' coverage quoted Eric Weinstein, managing director of Thiel Capital, who wrote that "destroying our own citizens sharing skepticism, doubt and information becomes 'Prebunking non-state malinformation.'" Separately, RT, the Russian state broadcaster, covered the 2022 refugee campaign under the headline "Google to 'inoculate' Eastern Europeans against disinformation", putting "inoculate" in quotation marks.

What the evidence supports and doesn't

Supported: short games and videos raise people's ability to name a manipulation technique in a test soon afterwards, in many countries, and campaigns can reach tens of millions of viewers.
Disputed: whether this is better discernment or a general tilt towards calling everything false. Independent reanalyses of Bad News say mostly the latter; the developers' signal detection meta-analysis says the former.
Not shown: that prebunking changes what people believe or share in real feeds over weeks, or that any campaign reduced the spread of the narratives it targeted. Campaign results come mainly from company surveys.
Not settled by research: who should choose the narratives to prebunk, a political question argued by the named critics above.

Images

Image	Source	Licence
Sander van der Linden, Hay Festival, 2019	Wikimedia Commons, LiteraryCritic1	CC BY-SA 4.0
Sind wir die Barbaren? (1916), Remember Belgium (1915), Spartakus bei der Arbeit (1919)	Gallery objects; provenance in Sind wir die Barbaren? (Are We the Barbarians?), Remember Belgium — Enlist To-Day and Spartakus bei der Arbeit (Spartacus at Work)	See each entry

Image gaps

Photos of McGuire or Roozenbeek. None found on Wikimedia Commons under an open licence.
Screenshots of the games or Jigsaw videos. Not included: they are copyrighted works and no openly licensed copy was found.

Sources

Techniques

Scapegoating Whataboutism Fear appeal

Inoculation and prebunking