Theme

“Honesty and inoculation: why every argument here gets its full strength”

Isn't it dangerous to state propaganda arguments in their strongest form? Doesn't that spread them? This page explains why this site does it on purpose, what research says about it, and what keeps it honest.

  1. The picture

    Why we state every argument at full strength.

    A persuasive argument works best alone: the image, the number, the slogan, with nothing beside it. Silencing it doesn't help, because silence feeds 'they're hiding something'. What takes away its pull is completing it: stating it the way its best advocates would, then showing what's true in it, what it leaves out, where it leads, and which reasoning trap it tends to fall into. John Stuart Mill put the reason in 1859: 'He who knows only his own side of the case, knows little of that.' Someone who has seen the whole of an argument and still disagrees can't be dismissed with 'you don't know the full picture', and neither can their opponent.

    primary documentJohn Stuart Mill, On Liberty (1859), chapter 2 (1859) ✓

    Chapter II, 'Of the Liberty of Thought and Discussion'; the passage 'He who knows only his own side of the case, knows little of that …' read in the Project Gutenberg text. Re-read 2026-10-12 (Gutenberg text): 'He must be able to hear them from persons who actually believe them; who defend them in earnest, and do their very utmost for them. He must know them in their most plausible and persuasive form'.

    Open source ↗

    What the research says: inoculation works.

    In the 1960s the psychologist William McGuire showed that people exposed to a weakened version of a persuasive attack, together with its refutation, resist the full attack better later, like a vaccine. Recent 'prebunking' research at Cambridge, by Sander van der Linden, Jon Roozenbeek and colleagues, found that teaching people the techniques of manipulation (emotional language, false dichotomies, scapegoating) protects them across topics, and a 2021 study by Rakoen Maertens and colleagues found that the protection lasted at least three months when people were tested at regular intervals, but decayed significantly within two months without that repetition — so it needs refreshing. The feared 'backfire effect', that corrections make people believe falsehoods more, has proved hard to reproduce: in five experiments with more than 10,100 subjects and 52 issues, Thomas Wood and Ethan Porter 'found no corrections capable of triggering backfire', and concluded that citizens 'by and large' heed factual information even when it challenges their ideological commitments ('The Elusive Backfire Effect', Political Behavior, published online 2018; the research was supported by the Cato Institute). 'By and large' is their own hedge: corrections not backfiring does not mean they always persuade.

    dataRakoen Maertens, Jon Roozenbeek, Melisa Basol and Sander van der Linden, 'Long-term effectiveness of inoculation against misinformation: Three longitudinal experiments', Journal of Experimental Psychology: Applied (2021) (2021) finding ✓

    Abstract read via the Europe PMC REST API (PubMed serves a reCAPTCHA): Bad News game; effect stable for at least 3 months with regular testing (Exp. 1), significant decay over 2 months without it (Exp. 2). Re-read 2026-10-11: participants 'develop resistance against misinformation through exposure to weakened doses of misinformation techniques'; outcome: ratings of headlines' reliability ('participants rate fake news as significantly less reliable after the intervention'); samples of 151, 194 and 170 against a Tetris control.

    Open source ↗

    dataThomas Wood and Ethan Porter, 'The Elusive Backfire Effect: Mass Attitudes' Steadfast Factual Adherence', Political Behavior (online 16 January 2018; print 2019) (2018) finding ✓

    Abstract read on Springer Nature Link (29 Sept 2026): five experiments, 'more than 10,100 subjects', '52 issues of potential backfire'; 'we found no corrections capable of triggering backfire'; 'Evidence of factual backfire is far more tenuous than prior research suggests. By and large, citizens heed factual information, even when such information challenges their ideological commitments.' It answers Nyhan and Reifler (2010). Body paywalled. The acknowledgements credit research support from the Cato Institute. A duplicate card (wood-porter-2019) was merged into this one.

    Open source ↗

    How this site's format is an inoculation.

    The toolkit cards name the technique, the weakened dose, and they apply to both sides and to topics this site never covers. 'What's omitted' and 'Where it leads' answer the argument before it lands. And a reference that people return to, rather than a one-time debunk, is the booster.

    No fear of the strongest version.

    Much persuasion in this conflict runs on fear: of erasure, of betrayal, of the other side's worst intentions. Fear works best when the threat stays vague and can't be examined. Stating the worst argument openly, calmly and at full strength removes that fuel: once it has been said and looked at, it has nothing left to threaten with. The research on fear appeals does not simply back this: a 2015 meta-analysis of 127 articles found that fear appeals do persuade (average effect d = 0.29), and more so when the message includes efficacy statements, that is, ways to act on the threat (Tannenbaum and colleagues, Psychological Bulletin). The same literature describes the other branch: 'strong fear appeals with low-efficacy messages produce the greatest levels of defensive responses', such as 'defensive avoidance or reactance' (Witte and Allen, Health Education & Behavior, 2000), and a re-analysis of the strictest experiments found threat changed behaviour 'only … under high efficacy' (Peters, Ruiter and Kok, Health Psychology Review). On this site's reading, a charge that no action could answer — 'stop the genocide' put to Israelis who see nothing they could do or stop doing that would end the argument, or a charge aimed the other way that no Palestinian act could meet — leaves its audience only that defensive branch, and the reactions can turn more extreme. That calm examination takes the force out of a fear appeal is this site's own wager, not a finding sourced here ⚠.

    dataMelanie B. Tannenbaum, Justin Hepler, Rick S. Zimmerman et al., 'Appealing to fear: A meta-analysis of fear appeal effectiveness and theories', Psychological Bulletin 141(6), 2015, pp. 1178–1204 (2015) finding ✓

    Abstract read via OpenAlex, 2026-10-02. 127 articles, 248 independent samples, N = 27,372. ‘Results showed a positive effect of fear appeals on attitudes, intentions, and behaviors’, random-effects d = 0.29; effectiveness ‘increased when the message included efficacy statements, depicted high susceptibility and severity, recommended one-time only (vs. repeated) behaviors’, and with more female recipients; conclusion: fear appeals ‘are effective at positively influencing attitude, intentions, and behaviors’.

    Open source ↗

    dataKim Witte and Mike Allen, 'A Meta-Analysis of Fear Appeals: Implications for Effective Public Health Campaigns', Health Education & Behavior 27(5), 2000, pp. 591–615 (2000) finding ✓

    Abstract read via OpenAlex, 2026-10-02. Strong fear appeals 'are more persuasive than low or weak fear appeals'; fear appeals motivate both 'adaptive danger control actions such as message acceptance and maladaptive fear control actions such as defensive avoidance or reactance'; 'strong fear appeals and high-efficacy messages produce the greatest behavior change, whereas strong fear appeals with low-efficacy messages produce the greatest levels of defensive responses'.

    Open source ↗

    dataGjalt-Jorn Y. Peters, Robert A. C. Ruiter and Gerjo Kok, 'Threatening communication: a critical re-analysis and a revised meta-analytic test of fear appeal theory', Health Psychology Review 7(sup1), pp. S8–S31 (2013) finding ✓

    Abstract read via OpenAlex, 2026-10-02. Six studies with full factorial manipulations of threat and efficacy and behaviour as outcome: 'threat only had an effect under high efficacy (d = 0.31), and efficacy only had an effect under high threat (d = 0.71)'; 'Threatening communication should exclusively be used when pilot studies indicate that an intervention successfully enhances efficacy.'

    Open source ↗

    Why honesty is the precondition.

    Inoculation only works if the reader trusts the dose. A debunk that concedes nothing reads as propaganda from the other side. That's why each entry grants what is true in the argument, in the site's own voice, before it shows what's missing; why the toolkit cards catch both sides' versions of each trap; and why the site shows its own workings: verification numbers generated from its database (including how few sources have been checked against the original so far), a description of the biases of the AI model that helps write it, public corrections, and a statement of where its maker stands (see 'This site is propaganda').

  2. Where it leads if pressed

    Working it out: what the research supports, and what this site assumes

    The question: does stating every argument at full strength, then answering it, protect readers, as this page argues? What the evidence establishes: in the 2021 Cambridge experiments, players of the Bad News game were exposed to 'weakened doses of misinformation techniques' and afterwards rated fake headlines 'significantly less reliable' than a control group did; the effect lasted at least three months with regular testing and faded within two months without it (samples of 151, 194 and 170). Across 52 issues, Wood and Porter found that factual corrections did not backfire 'by and large'. Fear appeals persuade on average (d = 0.29), and produce defensive responses when the audience is given nothing it can do. The page's inferences, tested against that. (1) 'Inoculation works': supported for a short game that teaches techniques, measured by how people rate headlines. Long pages on one conflict were not what was tested. The toolkit cards, which name techniques and apply to both sides, are the part of this site closest to what the experiments used, and one preregistered study comes closer still: a single video explaining the shared 'rhetorical markers' of Islamophobic and radical-Islamist disinformation, shown 'in a neutral context', left viewers agreeing less with 'gateway' videos from both directions, rating them less reliable and less willing to share them (Lewandowsky and Yesilada, 2021). That is argument-based inoculation against both sides of a charged divide, still a short video rather than long pages. (2) 'Full strength' against 'weakened dose': the experiments used weakened doses; this site states arguments at their strongest. The research cited here does not test whether a steelman followed by an answer protects a reader or persuades them; a reader could come away convinced by the steelman and unmoved by the answer. (3) 'A reference people return to is the booster': this depends on readers coming back, and no figure for returning readers is sourced here. (4) 'Corrections don't backfire': Wood and Porter tested factual corrections, and hedged their own result; arguments about identity and war may behave differently, and their study does not say. What's at stake: for critics on either side, a site that states the other side's case at full strength may be spreading it; for this site, its method rests on research that supports a narrower thing than the method does. Holds on the evidence here: teaching manipulation techniques measurably improves how people judge misleading headlines, for months if refreshed, and how they react to extremist videos from both sides of one divide, and factual corrections usually do not backfire. Open until a test shows it: that this site's pages protect readers, or that a full-strength steelman with its answer inoculates rather than persuades. Not supported: that the research shows this site works, or that stating an argument at full strength is harmless by default. A before-and-after measure such as the one under 'Limits' is the test that would move the middle line.

    What the labels (Holds, Open, Arguable …) mean

    dataRakoen Maertens, Jon Roozenbeek, Melisa Basol and Sander van der Linden, 'Long-term effectiveness of inoculation against misinformation: Three longitudinal experiments', Journal of Experimental Psychology: Applied (2021) (2021) finding ✓

    Abstract read via the Europe PMC REST API (PubMed serves a reCAPTCHA): Bad News game; effect stable for at least 3 months with regular testing (Exp. 1), significant decay over 2 months without it (Exp. 2). Re-read 2026-10-11: participants 'develop resistance against misinformation through exposure to weakened doses of misinformation techniques'; outcome: ratings of headlines' reliability ('participants rate fake news as significantly less reliable after the intervention'); samples of 151, 194 and 170 against a Tetris control.

    Open source ↗

    dataThomas Wood and Ethan Porter, 'The Elusive Backfire Effect: Mass Attitudes' Steadfast Factual Adherence', Political Behavior (online 16 January 2018; print 2019) (2018) finding ✓

    Abstract read on Springer Nature Link (29 Sept 2026): five experiments, 'more than 10,100 subjects', '52 issues of potential backfire'; 'we found no corrections capable of triggering backfire'; 'Evidence of factual backfire is far more tenuous than prior research suggests. By and large, citizens heed factual information, even when such information challenges their ideological commitments.' It answers Nyhan and Reifler (2010). Body paywalled. The acknowledgements credit research support from the Cato Institute. A duplicate card (wood-porter-2019) was merged into this one.

    Open source ↗

    dataMelanie B. Tannenbaum, Justin Hepler, Rick S. Zimmerman et al., 'Appealing to fear: A meta-analysis of fear appeal effectiveness and theories', Psychological Bulletin 141(6), 2015, pp. 1178–1204 (2015) finding ✓

    Abstract read via OpenAlex, 2026-10-02. 127 articles, 248 independent samples, N = 27,372. ‘Results showed a positive effect of fear appeals on attitudes, intentions, and behaviors’, random-effects d = 0.29; effectiveness ‘increased when the message included efficacy statements, depicted high susceptibility and severity, recommended one-time only (vs. repeated) behaviors’, and with more female recipients; conclusion: fear appeals ‘are effective at positively influencing attitude, intentions, and behaviors’.

    Open source ↗

    dataKim Witte and Mike Allen, 'A Meta-Analysis of Fear Appeals: Implications for Effective Public Health Campaigns', Health Education & Behavior 27(5), 2000, pp. 591–615 (2000) finding ✓

    Abstract read via OpenAlex, 2026-10-02. Strong fear appeals 'are more persuasive than low or weak fear appeals'; fear appeals motivate both 'adaptive danger control actions such as message acceptance and maladaptive fear control actions such as defensive avoidance or reactance'; 'strong fear appeals and high-efficacy messages produce the greatest behavior change, whereas strong fear appeals with low-efficacy messages produce the greatest levels of defensive responses'.

    Open source ↗

    referenceStephan Lewandowsky and Muhsin Yesilada, 'Inoculating against the spread of Islamophobic and radical-Islamist disinformation', Cognitive Research: Principles and Implications (2021), doi:10.1186/s41235-021-00323-z (2021) ✓

    Abstract read via OpenAlex (2026-10-12). Preregistered study; the inoculation group 'watched a video that explained common rhetorical markers of radical-Islamist and Islamophobic disinformation', presented 'in a neutral context not involving Islam'; then target videos with 'gateway' content for 'potential Islamist or Islamophobic radicalization'. Inoculated participants 'displayed less agreement with the video content, perceived the video as less reliable, and were less likely to share it'; 'support for the use of argument-based inoculation in combatting extremist messages'. Open access; only the abstract is read here.

    Open source ↗

    Limits.

    There is no evidence yet that this site works. Inoculation research tests short interventions on specific techniques; whether long argument pages on a conflict this charged move anyone is an open question, and it could be tested, for example with a misinformation-susceptibility measure before and after reading. Until then, this page is a reason for the method, not proof of its effect.

    dataRakoen Maertens, Friedrich M. Götz, Hudson Golino, Jon Roozenbeek, Claudia R. Schneider et al., 'The Misinformation Susceptibility Test (MIST): A psychometrically validated measure of news veracity discernment', Behavior Research Methods (2023) (2023) finding ✓

    Abstract read via OpenAlex (2026-09-30; open access). Three studies, seven samples, N = 8,504; the MIST-20, MIST-16 and MIST-8 ('completion time < 2 minutes'); validated in 'five national quota samples (US, UK), across 2 years'; the authors propose it 'as a screening tool, covariate, and intervention evaluation framework'. Supports the page's suggestion of a before-and-after measure; it says nothing about this site.

    Open source ↗

Keep exploring