CoChat
Sign Up
Cited by 514Misinformation and Its ImpactsSpam and Phishing DetectionIslamic Finance and Communication

Psychological inoculation improves resilience against misinformation on social media

Open access
Roozenbeek, Jon, van der Linden, Sander, Goldberg, Beth, Rathje, Steve, Lewandowsky, Stephan·2022·Science Advances
DOI 10.1126/sciadv.abo6254PMC PMC9401631
Open this paper in CoChat — free account

Abstract

Online misinformation continues to have adverse consequences for society. Inoculation theory has been put forward as a way to reduce susceptibility to misinformation by informing people about how they might be misinformed, but its scalability has been elusive both at a theoretical level and a practical level. We developed five short videos that inoculate people against manipulation techniques commonly used in misinformation: emotionally manipulative language, incoherence, false dichotomies, scapegoating, and ad hominem attacks. In seven preregistered studies, i.e., six randomized controlled studies ( n = 6464) and an ecologically valid field study on YouTube ( n = 22,632), we find that these videos improve manipulation technique recognition, boost confidence in spotting these techniques, increase people’s ability to discern trustworthy from untrustworthy content, and improve the quality of their sharing decisions. These effects are robust across the political spectrum and a wide variety of covariates. We show that psychological inoculation campaigns on social media are effective at improving misinformation resilience at scale.

AI snapshot

Written from the paper’s full text.

TL;DR

Five short inoculation videos targeting manipulation techniques improved recognition, confidence, trustworthiness discernment, and sharing decisions in six randomized studies (n = 6464) and a YouTube field study (n = 22,632). Effects were robust across the political spectrum and covariates, though the YouTube field effect narrowly missed the preregistered smallest effect size of interest.

Key findings

  • In studies 1–5, 16 of 20 preregistered outcome measures were supported; technique-recognition effect sizes ranged from Cohen’s d = 0.28 for scapegoating to d = 0.68 for false dichotomies. Nonsignificant effects were confidence and sharing for incoherence (study 2) and trustworthiness and sharing for scapegoating (study 4).
  • Study 6 replicated the emotional-language video: technique recognition P < 0.001, d = 0.67; trustworthiness P < 0.001, d = 0.44; sharing P < 0.001, d = 0.34. Outcome-measure response order did not significantly interact with condition (all P > 0.351).
  • In the YouTube field study (study 7), all six headlines combined showed significantly higher correct identification in treatment vs. control (P < 0.001, Cohen’s h = 0.09); emotional language P = 0.009, h = 0.05; false dichotomies P < 0.001, h = 0.13. The overall effect narrowly failed the preregistered SESOI of h = 0.10.
  • Moderation analyses found no consistently significant two-way interactions between experimental condition and covariates including gender, age, education, political ideology, news checking, social media use, populism, bullshit receptivity, conspiracy belief, analytical thinking, numeracy, personality, actively open-minded thinking, and misinformation susceptibility.

Methods

  • Seven preregistered studies: five randomized controlled studies testing five videos (total n = 5416, U.S. national quota samples), a replication of the emotional-language study (study 6, n = 1068), and an ecologically valid YouTube field study (study 7, n = 22,632).
  • In studies 1–6, participants were randomly assigned to watch a 1.5-min inoculation video or a neutral control video, then rated 10 synthetic social media posts (manipulative or neutral) on technique recognition, confidence, trustworthiness, and willingness to share; main outcomes used discernment scores (manipulative vs. neutral difference).
  • Study 7 ran two videos as YouTube ads; a random 30% of viewers were shown one test question within 24 hours, and a matched control group answered the same question without seeing an inoculation video. Analyses included independent-samples t tests, Bayesian analyses, moderation regressions, and two-proportion z tests.

Limitations

  • The paper states: the field study could not assess how long inoculation effects remain significant beyond the 24-hour window; videos and samples were tailored to U.S. audiences, so transfer to other cultural and linguistic settings is unknown; cross-protection against non-inoculated techniques was not investigated; manipulative and neutral stimuli were not exact mirrors; YouTube participants each saw only a single test item; and the YouTube study assessed susceptibility, not behavior such as sharing or watch time.
  • Evident from the text: laboratory sharing outcomes were self-reported willingness to share rather than observed sharing behavior, and the YouTube field effect was small (h = 0.09) and missed the preregistered SESOI overall.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1126/sciadv.abo6254

Cites

55 of 57 references verified1 citation errors · 1 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 37 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    In studies 1 to 5, the preregistered hypotheses were supported for 16 of 20 outcome measures.

    “Figure 2 and Table 1 show that our preregistered hypotheses (see Materials and Methods) are supported for 16 of 20 outcome measures, except for the confidence and sharing measures for the incoherence video (study 2) and the trustworthiness and sharing measures for the scapegoating video (study 4)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.
  • Findingown data

    Watching an inoculation video increases people's confidence in their ability to recognize manipulation techniques.

    “We thus find that watching an inoculation video improves people’s ability to recognize manipulation techniques in social media content and increases their confidence in their ability to do so.”
    • Also stated in: § Abstract · plain
      “boost confidence in spotting these techniques”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.
  • Findingown data

    In study 6, participants in the inoculation group have significantly higher discernment than the control group for trustworthiness (P < 0.001, d = 0.44).

    “We replicate the results from study 1: Participants in the inoculation group have significantly higher discernment than the control group for technique recognition (P < 0.001, d = 0.67), trustworthiness (P < 0.001, d = 0.44), and sharing (P < 0.001, d = 0.34)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.
  • Findingown data

    In study 7, the false dichotomies video significantly increased correct answers (P < 0.001, h = 0.13).

    “This is true for both the emotional language video (P = 0.009, h = 0.05) and for the false dichotomies video (P < 0.001, h = 0.13).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.
    • Rests on an assumption: The headlines used in study 7 contain a particular manipulation technique.
  • Findingown data

    There are no consistently significant two-way interactions between the experimental condition and each of the covariates in studies 1 to 6.

    “We find that there are no consistently significant two-way interactions between the experimental condition and each of the covariates.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.
  • May combine two claims. 21
  • Classified as probably not a substantial claim. 38
  • The quote only partly states this claim. 1

  • May combine two claims.

    Findingown data

    When considering each video on its own, the SESOI was not met for the emotional language video but was met for the false dichotomies video.

    “When considering each video on its own, the SESOI was not met for the emotional language video but it was met for the false dichotomies video”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.
    • Rests on an assumption: The headlines used in study 7 contain a particular manipulation technique.

  • May combine two claims.

    Findingown data

    Overall in study 7, although significant effects were found for both videos, the observed effect size across all six items (h = 0.09) narrowly failed to reach the preregistered smallest effect size of interest (SESOI) of h = 0.10.

    “Overall, although we find significant effects for both videos, we narrowly failed to reach our preregistered smallest effect size of interest (SESOI) of h = 0.10, with an observed effect size of h = 0.09 across all six items.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.
    • Rests on an assumption: The headlines used in study 7 contain a particular manipulation technique.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    A total of around 967,000 YouTube users watched one of the inoculation videos.

    “As part of this study, a total of around 967,000 YouTube users watched one of the inoculation videos.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.
    • Rests on an assumption: The headlines used in study 7 contain a particular manipulation technique.

  • May combine two claims.

    Findingown data

    The videos improve the quality of people's sharing decisions, i.e., they are either less likely to share manipulative content or more likely to share neutral/non-manipulative content.

    “In addition, the videos improve people’s ability to distinguish trustworthy and untrustworthy content, as well as the quality of their sharing decisions (i.e., they are either less likely to share manipulative content with others or more likely to share neutral/non-manipulative content).”
    • Also stated in: § Abstract · plain
      “improve the quality of their sharing decisions”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.

  • May combine two claims.

    Findingown data

    In study 6, discernment is significantly higher in the inoculation condition than the control condition for all three outcome measures when looking at each of the three response orders individually (all P values < 0.003).

    “Furthermore, discernment is significantly higher in the inoculation condition than the control condition for all three outcome measures when looking at each of the three response orders individually (all P values < 0.003; see table S51)”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.

  • May combine two claims.

    Findingown data

    The results were replicated in an ecologically valid setting on YouTube, boosting manipulation technique recognition by about 5% on average.

    “Crucially, we replicated these results in an ecologically valid setting on YouTube, boosting manipulation technique recognition by about 5% on average”
    Quote in the paperSubstantial claim
    • Rests on an assumption: YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.
    • Rests on an assumption: The headlines used in study 7 contain a particular manipulation technique.

  • May combine two claims.

    Findingown data

    Nineteen of 23 hypothesized effects from studies 1 to 6 were significant, and three of the four effects that were not significant were in the hypothesized direction.

    “Nineteen of 23 hypothesized effects from studies 1 to 6 were significant, and three of the four effects that were not significant were in the hypothesized direction.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.

  • May combine two claims.

    Findingown data

    The field study found that the videos were effective within a 24-hour window, with a median time of 18.4 hours between watching the video and responding to the survey question.

    “although our field study found that the videos were effective within a 24-hour window (with a median time of 18.4 hours between watching the video and responding to the survey question)”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.
    • Rests on an assumption: The headlines used in study 7 contain a particular manipulation technique.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Over a period of 15 days, the videos were shown to about 5.4 million people, or around 350,000 per day.

    “over a period of 15 days, the videos were shown to about 5.4 million people, or around 350,000 per day (see table S56).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In total, 22,632 participants (11,432 treatment and 11,200 control) answered a test question in study 7.

    “In total, 22,632 participants (11,432 treatment and 11,200 control) answered a test question.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Over a period of 15 days, the YouTube ad campaign videos were shown to about 5.4 million YouTube users.

    “Over this period, the videos were shown to about 5.4 million YouTube users.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    A total of 967,347 people watched either the emotional language video (422,371 views) or the false dichotomies video (544,976 views) as a YouTube ad for 30 s or more.

    “A total of 967,347 people watched either the emotional language video (422,371 views) or the false dichotomies video (544,976 views) as a YouTube ad for 30 s or more; see table S56.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

    Findingown data

    The average cost per 30-s video view in study 7 was $0.05.

    “The average cost per 30-s video view was $0.05.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Premisecitation

    Establishing what counts as factual information is epistemologically difficult, particularly in the context of politics.

    “Establishing what counts as factual information is epistemologically difficult, particularly in the context of politics (9);”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] S. Coleman, The elusiveness of political truth: From the conceit of objectivity to intersubjective judgement. Eur. J. Commun. 33, 157– 171 (2018). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Debunking misinformation is problematic because correcting misinformation does not always nullify its effects entirely, a phenomenon known as the continued influence effect.

    “Debunking misinformation is also problematic because correcting misinformation does not always nullify its effects entirely, a phenomenon known as the “continued influence effect” (13).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [13] S. Lewandowsky, U. K. H. Ecker, C. M. Seifert, N. Schwarz, J. Cook, Misinformation and its correction. Psychol. Sci. Public Interes. 13, 106– 131 (2012). 26173286 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Belief in misinformation about COVID-19 is linked to reduced willingness to get vaccinated and lower intentions to comply with public health measures.

    “belief in misinformation about coronavirus disease 2019 (COVID-19) has been linked to reduced willingness to get vaccinated against the disease and lower intentions to comply with public health measures (2, 3).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [2] J. Roozenbeek, C. R. Schneider, S. Dryhurst, J. Kerr, A. L. J. Freeman, G. Recchia, A. M. van der Bles, S. van der Linden, Susceptibility to misinformation about COVID-19 around the world. R. Soc. Open Sci. 7, 201199 (2020). 33204475 PMC7657933 source

      Not checked in the free run.

    • cites [3] S. Loomba, A. de Figueiredo, S. J. Piatek, K. de Graaf, H. J. Larson, Measuring the impact of COVID-19 vaccine misinformation on vaccination intent in the UK and USA. Nat. Hum. Behav. 5, 337– 348 (2021). 33547453 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Inoculation theory posits that it is possible to build psychological resistance against unwanted persuasion attempts, much like medical inoculations build physiological resistance against pathogens.

    “Inoculation theory follows a medical immunization analogy and posits that it is possible to build psychological resistance against unwanted persuasion attempts, much like medical inoculations build physiological resistance against pathogens.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [14] W. J. McGuire, D. Papageorgis, The relative efficacy of various types of prior belief-defense in producing immunity against persuasion. J. Abnorm. Soc. Psychol. 62, 327– 337 (1961). 13773992 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Prebunking has gained prominence as a means to preemptively build resilience against anticipated exposure to misinformation.

    “prebunking has gained prominence as a means to preemptively build resilience against anticipated exposure to misinformation (4).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [4] J. Roozenbeek, S. van der Linden, Fake news game confers psychological resistance against online misinformation. Humanit. Soc. Sci. Commun. 5, 1– 10 (2019).

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Premisecitation

    Testing the effectiveness of interventions in the real world, as opposed to a laboratory setting, is complicated.

    “testing the effectiveness of interventions in the real world (as opposed to a laboratory setting) is complicated (6).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [6] G. Pennycook, Z. Epstein, M. Mosleh, A. Arechar, D. Eckles, D. G. Rand, Shifting attention to accuracy can reduce misinformation online. Nature 592, 590– 595 (2021). 33731933 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Online misinformation is an important societal problem.

    “Online misinformation continues to have adverse consequences for society.”
    • Also stated in: § INTRODUCTION · plain
      “Online misinformation is an important societal problem (1).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] S. Lewandowsky, U. K. H. Ecker, J. Cook, Beyond misinformation: Understanding and coping with the “Post-Truth” era. J. Appl. Res. Mem. Cogn. 6, 353– 369 (2017). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    The five manipulation techniques used in the videos are consistently identified as epistemologically dubious: emotionally manipulative rhetoric, incoherent or mutually exclusive arguments, false dichotomies or dilemmas, scapegoating individuals or groups, and ad hominem attacks.

    “These five techniques were taken from the broader literature on argumentation and manipulation strategies and are consistently identified as epistemologically dubious: (i) using emotionally manipulative rhetoric to evoke outrage, anger, or other strong emotions (18, 22), (ii) the use of incoherent or mutually exclusive arguments (23), (iii) presenting false dichotomies or dilemmas (24), (iv) scape”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [18] C. Carrasco-Farré, The fingerprints of misinformation: How deceptive content differs from reliable sources in terms of cognitive effort and appeal to emotions. Humanit. Soc. Sci. Commun. 9, 162 (2022). source

      Not checked in the free run.

    • cites [22] W. J. Brady, J. A. Wills, J. T. Jost, J. A. Tucker, J. J. Van Bavel, Emotion shapes the diffusion of moralized content in social networks. Proc. Natl. Acad. Sci. U.S.A. 114, 7313– 7318 (2017). 28652356 PMC5514704 source

      Not checked in the free run.

    • cites [23] S. Lewandowsky, J. Cook, E. A. Lloyd, The “Alice in Wonderland” mechanics of the rejection of (climate) science: Simulating coherence by conspiracism. Synthese 195, 175– 196 (2016). source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Traditional inoculation research has focused on building resilience against specific persuasive attacks.

    “traditional inoculation research has focused on building resilience against specific persuasive attacks (15).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15] J. Compton, in The SAGE Handbook of Persuasion: Developments in Theory and Practice, J. P. Dillard, L. Shen, Eds. (SAGE Publications Inc., ed. 2, 2013), pp. 220–236. source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Each social media post was randomly either manipulative or a neutral counterpart.

    “Each post was randomly either manipulative (i.e., it made use of a manipulation technique) or a neutral counterpart (similar in content and length to the manipulative post but not making use of the manipulation technique).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Psychological inoculation treatments contain two core components: a forewarning that induces a perceived threat of an impending attack on one's attitudes and exposure to a weakened (micro)dose of misinformation that contains a preemptive refutation (or prebunk) of the anticipated misleading arguments or persuasion techniques.

    “Psychological inoculation treatments contain two core components (15): (i) a forewarning that induces a perceived threat of an impending attack on one’s attitudes and (ii) exposure to a weakened (micro)dose of misinformation that contains a preemptive refutation (or prebunk) of the anticipated misleading arguments or persuasion techniques.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15] J. Compton, in The SAGE Handbook of Persuasion: Developments in Theory and Practice, J. P. Dillard, L. Shen, Eds. (SAGE Publications Inc., ed. 2, 2013), pp. 220–236. source

      Not checked in the free run.

  • May combine two claims.

    Premisereasoning

    These tropes can be analyzed and used for inoculation without prior knowledge of specific misleading content, thus potentially providing broad resilience against social media or news content that draws on one or more of the techniques that someone has been inoculated against.

    “These tropes, which include manipulation techniques such as logical fallacies (20) and emotionally manipulative language (18), can be analyzed and used for inoculation without prior knowledge of specific misleading content, thus potentially providing broad resilience against social media or news content that draws on one or more of the techniques that someone has been inoculated against.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [20] E. K. Vraga, S. C. Kim, J. Cook, Testing logic-based and humor-based corrections for science, health, and political misinformation on social media. J. Broadcast. Electron. Media 63, 393– 414 (2019). source

      Not checked in the free run.

    • cites [18] C. Carrasco-Farré, The fingerprints of misinformation: How deceptive content differs from reliable sources in terms of cognitive effort and appeal to emotions. Humanit. Soc. Sci. Commun. 9, 162 (2022). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The seven studies were preregistered.

    “In seven preregistered studies”
    • Also stated in: § The present research · plain
      “We conducted seven preregistered studies to test the effectiveness of these videos as a way to inoculate people against online misinformation”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    In studies 1 to 6, participants were randomly assigned to watch either a 1.5-min inoculation video or a neutral control video of approximately equal length.

    “In studies 1 to 6, participants were randomly assigned to watch either a 1.5-min inoculation video or a neutral control video of approximately equal length.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S1 (asserted).

    Premisecitation

    How to reduce misinformation susceptibility in real-world environments where misinformation is regularly consumed remains an open question.

    “how to reduce misinformation susceptibility in real-world environments where misinformation is regularly consumed (such as social media and video sharing platforms) remains an open question (21).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [21] H. O.-Y. Li, A. Bailey, D. Huynh, J. Chan, YouTube as a source of information on COVID-19: A pandemic of misinformation? BMJ Glob. Heal. 5, e002604 (2020). PMC7228483 32409327 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Priming people to be more mindful of accuracy (accuracy nudges) was found to prompt Twitter users to share more high-quality sources such as CNN and the New York Times.

    “priming people to be more mindful of accuracy (through so-called accuracy nudges or accuracy prompts) was found to prompt Twitter users to share more high-quality sources such as CNN and the New York Times (6).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [6] G. Pennycook, Z. Epstein, M. Mosleh, A. Arechar, D. Eckles, D. G. Rand, Shifting attention to accuracy can reduce misinformation online. Nature 592, 590– 595 (2021). 33731933 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationown data

    The relative reduced effectiveness of the scapegoating and incoherence videos compared to the other three videos may be explained by variations in baseline discernment across techniques.

    “The relative reduced effectiveness of the scapegoating and incoherence videos compared to the other three videos may be explained by variations in baseline discernment across techniques (i.e., participants in the control group were better at identifying the scapegoating/incoherence technique than other techniques even without an intervention)”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The study was unable to investigate to what extent watching one inoculation video confers psychological resistance against manipulation techniques against which people were not specifically inoculated (cross-protection).

    “we were unable to investigate to what extent watching one inoculation video confers psychological resistance against manipulation techniques against which people were not specifically inoculated, a phenomenon known as “cross-protection” (48).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [48] K. A. Parker, S. A. Rains, B. Ivanov, Examining the “blanket of protection” conferred by inoculation: The effects of inoculation messages on the cross-protection of related attitudes. Commun. Monogr. 83, 49– 68 (2016). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Across seven preregistered studies including a field experiment on YouTube, with a total of nearly 30,000 participants, watching short inoculation videos improves people's ability to identify manipulation techniques commonly used in online misinformation, both in a laboratory setting and in a real-world environment where exposure to misinformation is common.

    “Across seven high-powered preregistered studies including a field experiment on YouTube, with a total of nearly 30,000 participants, we find that watching short inoculation videos improves people’s ability to identify manipulation techniques commonly used in online misinformation, both in a laboratory setting and in a real-world environment where exposure to misinformation is common.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.
    • Rests on an assumption: YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.
    • Rests on an assumption: The headlines used in study 7 contain a particular manipulation technique.

  • May combine two claims.

    Interpretationcitation

    The effect sizes are generally larger than those of existing scalable interventions; digital literacy tips improved detection of false headlines with a Cohen's d of 0.20.

    “These effect sizes are also generally larger than those of existing scalable interventions. For comparison, digital literacy tips (consisting of a short text pointing out various ways to spot false news, such as looking at the validity of the source) were found to improve detection of false headlines with a Cohen’s d of 0.20 (45).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [45] A. M. Guess, M. Lerner, B. Lyons, J. M. Montgomery, B. Nyhan, J. Reifler, N. Sircar, A digital media literacy intervention increases discernment between mainstream and false news in the United States and India. Proc. Natl. Acad. Sci. U.S.A. 117, 15536– 15545 (2020). 32571950 PMC7355018 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Participants in the YouTube field study each only saw a single test item.

    “participants in our YouTube field study each only saw a single test item, as it was only possible to embed a single question at the end of YouTube videos”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.

  • May combine two claims.

    Interpretationown data

    These interventions are effective for people with different ideological backgrounds and cognitive styles and are generalizable to a wide variety of key issue domains.

    “We show that these interventions are effective for people with different ideological backgrounds and cognitive styles and are generalizable to a wide variety of key issue domains.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.
    • Rests on an assumption: The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The YouTube field study only assessed misinformation susceptibility and not people's behavior, such as sharing misinformation or watch time.

    “our YouTube field study only assessed misinformation susceptibility and not people’s behavior (such as the sharing of misinformation with others or for how long people watch low-quality content on YouTube).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Both the videos and the study were tailored to U.S. audiences, and samples consisted of U.S. residents.

    “both the videos and the study were tailored to U.S. audiences, and our samples consisted of U.S. residents.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The field study confirms that inoculation videos could be run as public-service ads ahead of potentially harmful content and thereby easily scaled across millions of users.

    “Our field study thus confirms that inoculation videos could be run as public-service ads ahead of potentially harmful content and thereby easily scaled across millions of users.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.
    • Rests on an assumption: The headlines used in study 7 contain a particular manipulation technique.

  • Classified as probably not a substantial claim.

    Interpretationown data

    When optimizing for views (rather than survey responses) and with a larger budget, it is very doable to reach millions of people.

    “When optimizing for views (rather than survey responses) and with a larger budget, it is very doable to reach millions of people.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    For studies 1 to 6, all stimuli were reviewed by four experts.

    “All stimuli were reviewed by four experts and covered the use of manipulation techniques in contexts in which misinformation is common (such as climate change) but did not cover the same topics as the inoculation videos (see note x in the Supplementary Materials).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Each sample in studies 1 to 6 was nationally balanced on the U.S. population on age, gender, and ethnicity.

    “Each sample was nationally balanced on the U.S. population on age, gender, and ethnicity.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    For studies 1 to 5, the authors preregistered that they would first collect a pilot sample of n = 200 responses, then resume data collection for the remaining 884 target responses.

    “For studies 1 to 5, we preregistered that we would first collect a pilot sample of n = 200 responses (halting data collection at this point and analyzing the pilot data) to ensure that the survey was properly implemented and responses were collected correctly. After ensuring this, we resumed data collection for the remaining 884 target responses.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The studies used a randomized controlled design.

    “In seven preregistered studies (and one pilot study), we tested the efficacy of each of these videos using a randomized controlled design.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The efficacy of each inoculation video was tested in seven preregistered studies.

    “In seven preregistered studies (and one pilot study), we tested the efficacy of each of these videos using a randomized controlled design.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    For study 6, participants were randomly assigned to one of three outcome measure response orders (MTS, TSM, or SMT) after being randomly assigned to an experimental condition.

    “Participants were randomly assigned to one of three response orders (after being randomly assigned to an experimental condition): MTS (n = 364), TSM (n = 361), or SMT (n = 343).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    For the ad hominem study (study 5), an error in the approval process meant that the preregistration was approved by all authors after data were already collected, although the wording of the preregistration itself was not changed.

    “For the ad hominem study (study 5), an error in the approval process meant that the preregistration was approved by all authors after data were already collected; however, the wording of the preregistration itself was not changed.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductown data

    For the emotional language studies (studies 1 and 6), a stimulus validation test using sentiment analysis was conducted to confirm whether the manipulative stimuli capture the intended dimension of (manipulative) emotionality and the neutral stimuli do not.

    “We therefore conducted a stimulus validation test (using sentiment analysis) to confirm whether the manipulative stimuli capture the intended dimension of (manipulative) emotionality (and the neutral stimuli do not), for which we refer to Supplementary Analyses.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    In study 7, the authors were only able to implement one test question per participant because of limitations of the YouTube brand lift survey feature.

    “In study 7, because of the limitations of the YouTube brand lift survey feature, we were only able to implement one test question per participant.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    For studies 1 to 6, source information (profile picture, name, username, etc.) was blacked out for both manipulative and neutral posts to avoid biasing participants' responses.

    “For both the manipulative and neutral posts, source information (profile picture, name, username, etc.) was blacked out to avoid biasing participants’ responses.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    A pilot study (n = 194) was conducted with the emotional language video to validate the stimuli sets and outcome measures.

    “The pilot study (n = 194; see table S2 for the main results) was conducted with the emotional language video to validate our stimuli sets and outcome measures.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    In study 7, the authors only included test questions that contained a manipulation technique and no control/neutral items.

    “As we were limited in the number of test questions that we could administer because of the cost of running the YouTube campaign (see table S56) and were primarily interested in participants’ ability to identify manipulation techniques, we only included test questions that contained a manipulation technique and no control/neutral items.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    These videos are effective not only in a laboratory setting but also "in the wild" on a video sharing platform and can therefore be easily implemented at scale to improve resilience against misinformation at a cost of about $0.05 per video view.

    “We provide evidence that these videos are effective not only in a laboratory setting but also “in the wild” on a video sharing platform and can therefore be easily implemented at scale to improve resilience against misinformation at a cost of about $0.05 per video view.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.
    • Rests on an assumption: The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.
    • Rests on an assumption: The headlines used in study 7 contain a particular manipulation technique.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    For study 7, the authors ensured that there was no overlap between users for the different test questions.

    “We ensured that there was no overlap between users for the different test questions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Research has shown that populist attitudes, analytical thinking, numerical thinking, bullshit receptivity, conspiracy mentality, news checking frequency, and social media use are associated with susceptibility to misinformation.

    “we also included the following measures as covariates, which research has shown are associated with susceptibility to misinformation:”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [35] A. Akkerman, C. Mudde, A. Zaslove, How populist are the people? measuring populist attitudes in voters. Comp. Polit. Stud. 47, 1324– 1353 (2013). source

      Not checked in the free run.

    • cites [38] S. Frederick, Cognitive reflection and decision making. J. Econ. Perspect. 19, 25– 42 (2005). source

      Not checked in the free run.

    • cites [50] M. E. Toplak, R. F. West, K. E. Stanovich, The cognitive reflection test as a predictor of performance on heuristics-and-biases tasks. Mem. Cogn. 39, 1275– 1289 (2011). 21541821 source

      Not checked in the free run.

    • cites [36] G. Pennycook, J. A. Cheyne, N. Barr, D. J. Koehler, J. A. Fugelsang, On the reception and detection of pseudo-profound bullshit. Judgm. Decis. Mak. 10, 549– 563 (2015). source

      Not checked in the free run.

54 of 54

  • The paper relies on this without support.

    Nonpolitical and fictitious humorous examples avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.

    “All examples are nonpolitical and fictitious, in addition to being humorous, to avoid any appearance of partisan bias and prevent triggering defensive motivated cognition.”

  • The paper relies on this without support.

    The manipulative stimuli clearly make use of a specific manipulation technique, and the neutral counterparts do not.

    “The manipulative stimuli were not always necessarily false in the sense that they made verifiably untrue claims; rather, they were designed to clearly make use of a specific manipulation technique.”

  • The paper relies on this without support.

    YouTube users in the treatment and control groups are comparable except for exposure to the inoculation video.

    “a control group (a group of YouTube users matched to the treatment group in terms of demographics, who were not shown an inoculation video but did answer one of the test questions)”

  • The paper relies on this without support.

    The single-item test question embedded within YouTube is a valid measure of manipulation technique recognition.

    “a random 30% of this group was shown one single-item test question embedded within the YouTube environment, which consisted of a headline containing a particular manipulation technique”

  • The paper relies on this without support.

    The headlines used in study 7 contain a particular manipulation technique.

    “which consisted of a headline containing a particular manipulation technique (the headlines were adapted from the stimuli used in studies 1 and 3)”

  • The paper relies on this without support.

    The control video (a SciShow video about freezer burn) was of similar length and aesthetic to the inoculation videos and its content was unrelated to online misinformation.

    “the control group watched a video of similar length and aesthetic, but with content unrelated to online misinformation (specifically: a SciShow video about freezer burn, found here: www.youtube.com/watch?v=fPEtOaGTZ0s&t=8s).”

  • The paper relies on this without support.

    For studies 2 to 5 (incoherence, false dichotomies, scapegoating, and ad hominem), the manipulation techniques can be clearly defined and embedded in the stimuli.

    “For studies 2 to 5 (incoherence, false dichotomies, scapegoating, and ad hominem), stimuli design was straightforward, as these techniques can be clearly defined and embedded in the stimuli.”

  • The paper relies on this without support.

    The single test question embedded within YouTube in study 7 validly measures participants' ability to identify manipulation techniques.

    “Participants were shown a single sentence containing a manipulation technique (either emotional language or a false dichotomy; the stimuli themselves were taken from studies 1 and 3, with minor adaptations due to length limitations; see the “Stimuli” section below) and were asked to indicate what technique is used (four response options: one correct and three incorrect).”
8 of 8

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • The results support recent research showing that educational material on common misinformation techniques can improve people's ability to discern trustworthiness from untrustworthy videos.

    “These results support recent research showing that educational material on common misinformation techniques can improve people’s ability to discern trustworthiness from untrustworthy videos89.”

    From Human detection of political speech deepfakes across transcripts, audio, and video (2024) · cites this paper as [89]

    The Results (Studies 1 to 5) say that videos teaching common manipulation techniques improved participants’ ability to distinguish trustworthy from untrustworthy content. But the Introduction (The present research) specifies that participants judged synthetic social media posts, not the trustworthiness of videos. The videos were the educational intervention. (LLM, conf=0.88)

    Experimental — accuracy not yet measured

  • In a field study on YouTube, videos containing micro-doses of common misinformation techniques increased discernment of online manipulation tactics.

    “Similarly, in a field study on YouTube, videos containing micro-doses of common misinformation techniques increased discernment of online manipulation tactics (39).”

    From The impact of generative artificial intelligence on socioeconomic inequalities and policy making (2024) · cites this paper as [39]

Cites

Flagged first, then by citations

Research works printed but not linked: 2

  1. M. Mosleh C. Martel D. Eckles D. G. Rand in CHI ‘21: Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems (Yokohama Japan 8 to 13 May 2021) pp. 1–13.
  2. D. Walton Ad Hominem Arguments (The University of Alabama Press 1998).

Cited by

Sorted by citations

Related

Authors
Roozenbeek, J., van der Linden, S., Goldberg, B., Rathje, S., & Lewandowsky, S.
Affiliations
  • University of Cambridge (GB)
  • Department of Psychology, University of Cambridge, Cambridge, UK.
  • Google (United States) (US)
  • The University of Western Australia (AU)
Corresponding author
Roozenbeek, Jon
ORCID
Venue
Science Advances
Publisher
American Association for the Advancement of Science (AAAS)
Published
2022-08-26
Volume
8
Issue
34
Pages
eabo6254-eabo6254
ISSN
2375-2548 (ISSN-L)
DOI
10.1126/sciadv.abo6254
PMC
PMC9401631
OpenAlex
W4293105742
Semantic Scholar
67dc78b1f3cf4311fc30c31e5ee63ae3ba7c2379
Cited by
472
References
54
Citations by year
2026: 125 · 2025: 127 · 2024: 135 · 2023: 71 · 2022: 13 · 2020: 1
FWCI
124.83
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Science Advances (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Explore Bristol Research (publishedVersion, cc-by)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
  • UWA Profiles and Research Repository (UWA) (submittedVersion, cc-by)
  • OSF Preprints (OSF Preprints) (submittedVersion)
  • Apollo (cc-by)
Funders
Economic and Social Research Council (ES/V011960/1)
Topics
  • Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
  • Physical Sciences › Computer Science › Information Systems › Spam and Phishing Detection
  • Social Sciences › Sociology and Political Science › Islamic Finance and Communication
Keywords
psychological inoculation, misinformation, social media, manipulation techniques, emotional manipulation, false dichotomy, scapegoating, ad hominem, incoherence, manipulation detection, trustworthiness judgments, YouTube, political polarization, misinformation resilience

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review