arxiv: 2604.09200 · v1 · submitted 2026-04-10 · 💻 cs.CY · cs.AI· cs.HC

Recognition: unknown

Artificial intelligence can persuade people to take political actions

Kobi Hackenburg , Luke Hewitt , Caroline Wagner , Ben M. Tappin , Christopher Summerfield

Authors on Pith no claims yet

Pith reviewed 2026-05-10 16:46 UTC · model grok-4.3

classification 💻 cs.CY cs.AIcs.HC

keywords AI persuasionbehavioral outcomespolitical actionspetition signingattitude-behavior gapconversational AIinformation provisioncharitable donations

0 comments

The pith

Conversational AI persuades people to sign petitions and donate, but these effects do not track with attitude changes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether AI can move people beyond opinion shifts to actual actions such as signing real petitions or making donations. In two large preregistered experiments, conversational AI produced sizable increases in these behaviors. These behavioral effects showed no correlation with AI-driven changes in attitudes, and providing information that explained attitude shifts did not account for the action changes. Behavioral persuasion strategies all outperformed the strongest attitudinal strategy, with only small differences among them. The results indicate that studies limited to attitudes may not capture or may misrepresent AI's influence on what people actually do.

Core claim

In two large preregistered experiments (N=17,950 responses from 14,779 people), conversational AI models produced sizable effects on behavioral outcomes such as petition signing (+19.7 percentage points) and charitable donations, yet these effects showed no correlation with AI-induced attitude changes. Information provision drove attitude shifts but not behavioral ones. All eight tested behavioral persuasion strategies outperformed the most effective attitudinal strategy, though differences among the behavioral strategies were small.

What carries the argument

Experimental comparison of AI persuasion effects on paired attitudinal and behavioral measures, using conversational models to test real actions like petition signing and donations.

Load-bearing premise

That effects seen in an online setting with low-stakes real actions such as petition signing will reflect how AI persuades people in higher-stakes real-world decisions.

What would settle it

A follow-up study in which AI-induced attitude changes strongly predict the same participants' later behavioral changes, or a field experiment showing no increase in petition signing or donations from AI chatbots.

Figures

Figures reproduced from arXiv: 2604.09200 by Ben M. Tappin, Caroline Wagner, Christopher Summerfield, Kobi Hackenburg, Luke Hewitt.

**Figure 2.** Figure 2: AI conversations produce sizable persuasion effects on political behaviours. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: AI persuasion effects on attitudes and behaviour are uncorrelated and driven by [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: A combined persuasion strategy is most effective at driving behavioural change. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

read the original abstract

There is substantial concern about the ability of advanced artificial intelligence to influence people's behaviour. A rapidly growing body of research has found that AI can produce large persuasive effects on people's attitudes, but whether AI can persuade people to take consequential real-world actions has remained unclear. In two large preregistered experiments N=17,950 responses from 14,779 people), we used conversational AI models to persuade participants on a range of attitudinal and behavioural outcomes, including signing real petitions and donating money to charity. We found sizable AI persuasion effects on these behavioural outcomes (e.g. +19.7 percentage points on petition signing). However, we observed no evidence of a correlation between AI persuasion effects on attitudes and behaviour. Moreover, we replicated prior findings that information provision drove effects on attitudes, but found no such evidence for our behavioural outcomes. In a test of eight behavioural persuasion strategies, all outperformed the most effective attitudinal persuasion strategy, but differences among the eight were small. Taken together, these results suggest that previous findings relying on attitudinal outcomes may generalize poorly to behaviour, and therefore risk substantially mischaracterizing the real-world behavioural impact of AI persuasion.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Large preregistered experiments show AI shifting low-stakes online actions like petition signing by ~20 points with no attitude correlation, but the behaviors are too cheap to tell us much about real political influence.

read the letter

The punchline is that this paper gives evidence AI can move actual behaviors in online experiments—petition signing up 19.7 points, plus donations—while those shifts show no correlation with attitude changes and information provision fails to drive the behaviors the way it drives attitudes. All eight behavioral strategies they tested beat the strongest attitude one, with little difference among the eight. That dissociation is the clearest new claim and it directly addresses the gap they flag between prior attitude work and real actions. The scale helps: two preregistered studies, nearly 18,000 responses from almost 15,000 people, which is enough to make the null correlation and the strategy comparisons worth taking seriously on their own terms. They also replicate the attitude effect from information, so the contrast is clean. The soft spot is external validity. Petition signing and small online donations carry almost no cost or follow-through, so the sizable effects and the attitude-behavior split could be tied to the low-stakes online format rather than a general rule about how AI persuasion works. Demand characteristics are plausible here too—people might just comply with the chatbot because they are in an experiment. Without higher-stakes measures or field tests, it is hard to know whether the same strategies would move voting, protesting, or bigger donations. The paper is for people working on AI influence, computational social science, or behavioral interventions who already care about moving past survey attitudes. It is worth sending to peer review because the preregistration and sample size make the empirical results checkable, even if referees will rightly press on whether the behavioral measures are good proxies for consequential political action.

Referee Report

2 major / 3 minor

Summary. The paper reports two large preregistered experiments (N=17,950 responses from 14,779 participants) in which conversational AI models were used to persuade people on political topics. It finds sizable effects on behavioral outcomes such as real petition signing (+19.7 percentage points) and charitable donations, no evidence of correlation between AI persuasion effects on attitudes versus behaviors, replication of information provision driving attitudinal but not behavioral change, and that all eight tested behavioral persuasion strategies outperformed the strongest attitudinal strategy (with small differences among the behavioral ones). The authors conclude that prior attitudinal research may generalize poorly to behavior and thus risk mischaracterizing AI's real-world political impact.

Significance. If the results hold, the work is significant for providing direct evidence that AI can drive consequential political actions beyond attitude shifts. The preregistered design, large sample, and use of real (if low-stakes) behavioral measures are strengths that lend credibility. The dissociation between attitude and behavior effects, plus the uniform efficacy of behavioral strategies, challenges reliance on attitudinal proxies in AI persuasion studies and carries implications for understanding AI's role in political mobilization.

major comments (2)

The central interpretation (abstract and Discussion) that attitudinal findings 'risk substantially mischaracterizing the real-world behavioural impact' rests on the behavioral measures being valid proxies for consequential actions. The reported outcomes (online petition signing and small donations) involve negligible personal cost or follow-through; this raises the possibility that the observed dissociation from attitudes and the uniform superiority of the eight behavioral strategies reflect measurement artifacts or demand characteristics rather than a fundamental mechanistic difference. A direct test or sensitivity discussion of this assumption is needed to support the strong generalization claim.
Results section on the null correlation and information-provision findings: the manuscript reports 'no evidence of a correlation' and 'no such evidence' for information provision on behavior, but does not specify the exact statistical procedure (e.g., correlation across which units/conditions, regression specification, or power for the null). Because these nulls are load-bearing for the dissociation claim, the precise tests, effect-size estimates, and any multiple-comparison adjustments should be reported explicitly.

minor comments (3)

Abstract: the phrasing 'N=17,950 responses from 14,779 people' is ambiguous about whether participants could contribute multiple responses and how this affects the effective sample size per experiment; clarify in the main text as well.
Methods/Results: ensure all reported effect sizes (including the +19.7 pp example) are accompanied by confidence intervals and exact p-values to allow readers to assess precision.
Discussion: add a brief reference to existing attitude-behavior consistency literature (e.g., on low- vs. high-stakes actions) to better situate the dissociation finding.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful and constructive comments. We address each major comment below, indicating planned revisions where appropriate.

read point-by-point responses

Referee: The central interpretation (abstract and Discussion) that attitudinal findings 'risk substantially mischaracterizing the real-world behavioural impact' rests on the behavioral measures being valid proxies for consequential actions. The reported outcomes (online petition signing and small donations) involve negligible personal cost or follow-through; this raises the possibility that the observed dissociation from attitudes and the uniform superiority of the eight behavioral strategies reflect measurement artifacts or demand characteristics rather than a fundamental mechanistic difference. A direct test or sensitivity discussion of this assumption is needed to support the strong generalization claim.

Authors: We acknowledge that online petition signing and small donations entail lower personal costs than high-stakes actions such as voting or large financial commitments. These measures nonetheless capture real behavioral commitments with tangible downstream consequences, consistent with standard practices in behavioral science. The observed pattern—behavioral strategies outperforming attitudinal ones while information provision affects attitudes but not behavior—would be difficult to explain solely as demand characteristics, given the preregistered design and large sample. We will add an expanded limitations section with a sensitivity discussion of measurement assumptions, alternative interpretations, and tempered generalization language in the abstract and Discussion. revision: partial
Referee: Results section on the null correlation and information-provision findings: the manuscript reports 'no evidence of a correlation' and 'no such evidence' for information provision on behavior, but does not specify the exact statistical procedure (e.g., correlation across which units/conditions, regression specification, or power for the null). Because these nulls are load-bearing for the dissociation claim, the precise tests, effect-size estimates, and any multiple-comparison adjustments should be reported explicitly.

Authors: We agree that greater statistical transparency is required for the null results. In the revision we will specify: (1) the attitude-behavior correlation was computed as the Pearson correlation between condition-level mean attitude shifts and behavior rates; (2) information-provision effects on behavior were tested via OLS regression with the information condition as the key predictor, including demographic covariates; and (3) we will report effect sizes, 95% confidence intervals, and post-hoc power calculations for the null effects, along with any multiple-comparison corrections applied. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical experimental study with direct data comparisons

full rationale

This is a preregistered experimental paper reporting measured effects from two large-N studies on AI persuasion. All central claims rest on direct statistical comparisons between randomized conditions (e.g., AI vs. control on petition signing rates) and on replication of external prior findings. No equations, derivations, fitted parameters, or self-referential definitions appear; the behavioral outcomes are not constructed from the attitudinal measures or vice versa. Self-citations, if present, are limited to methodological precedents and do not bear the load of any uniqueness claim or ansatz. The design is therefore self-contained against external benchmarks of experimental validity.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper relies on standard assumptions of randomized controlled trials in psychology and no new entities or free parameters are introduced in the abstract.

axioms (2)

standard math Random assignment to conditions isolates the causal effect of AI persuasion
Core to experimental design in the studies.
domain assumption Self-reported attitudes and observed behaviors (petition signing, donations) are valid measures of persuasion outcomes
Assumed in behavioral science; if people lie or actions are not genuine, results misrepresent real impact.

pith-pipeline@v0.9.0 · 5511 in / 1497 out tokens · 62517 ms · 2026-05-10T16:46:27.176077+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

45 extracted references · 4 canonical work pages

[1]

Acting on intentions: The role of anticipated regret.British journal of social psychology, 42(4):495–511, 2003

Charles Abraham and Paschal Sheeran. Acting on intentions: The role of anticipated regret.British journal of social psychology, 42(4):495–511, 2003

2003
[2]

Evaluating language models for harmful manipulation, 2026

Canfer Akbulut, Rasmi Elasmar, Abhishek Roy, Anthony Payne, Priyanka Suresh, Lujain Ibrahim, Seliem El-Sayed, Charvi Rastogi, Ashyana Kachra, Will Hawkins, Kristian Lum, and Laura Weidinger. Evaluating language models for harmful manipulation, 2026

2026
[3]

L. P. Argyle, E. C. Busby, J. R. Gubler, A. Lyman, J. Olcott, J. Pond, and D. Wingate. Testing theories of political persuasion using ai.Proceedings of the National Academy of Sciences, 122:e2412815122, 2025

2025
[4]

Armitage and Mark Conner

Christopher J. Armitage and Mark Conner. Efficacy of the theory of planned behaviour: A meta-analytic review.British journal of social psychology, 40(4):471–499, 2001

2001
[5]

Llm-generated messages can persuade humans on policy issues.Nature Communications, 16(1):6037, 2025

Hui Bai, Jan G Voelkel, Shane Muldowney, Johannes C Eichstaedt, and Robb Willer. Llm-generated messages can persuade humans on policy issues.Nature Communications, 16(1):6037, 2025

2025
[6]

Daniel Batson.Altruism in humans

C. Daniel Batson.Altruism in humans. Oxford University Press, New York, 1 edition, 2011

2011
[7]

Brewer, Jessica T

Noel T. Brewer, Jessica T. DeFrank, and Melissa B. Gilkey. Anticipated regret and health behavior: A meta-analysis.Health psychology, 35(11):1264–1275, 2016

2016
[8]

helping" versus

Christopher J. Bryan, Allison Master, and Gregory M. Walton. "helping" versus "being a helper": Invoking the self to increase helping in young children.Child development, 85(5):1836–1842, 2014

2014
[9]

Motivating voter turnout by invoking the self.Proceedings of the National Academy of Sciences - PNAS, 108(31):12653–12656, 2011

Christopher J Bryan, Gregory M Walton, Todd Rogers, and Carol S Dweck. Motivating voter turnout by invoking the self.Proceedings of the National Academy of Sciences - PNAS, 108(31):12653–12656, 2011

2011
[10]

Burtell and T

M. Burtell and T. Woodside. Artificial influence: An analysis of ai-driven persuasion, 2023. arXiv:2303.08721 [Preprint]

work page arXiv 2023
[11]

Camilleri and Richard P

Adrian R. Camilleri and Richard P. Larrick. The collective aggregation effect: Aggregating potential collective action increases prosocial behavior.Journal of experimental psychology. General, 148(3):550– 569, 2019

2019
[12]

Z. Chen, J. Kalla, Q. Le, S. Nakamura-Sakai, J. Sekhon, and R. Wang. A framework to assess the persuasion risks large language model chatbots pose to democratic societies, 2025. arXiv:2505.00036 [Preprint]

work page arXiv 2025
[13]

Cialdini and Noah J

Robert B. Cialdini and Noah J. Goldstein. Social influence: Compliance and conformity.Annual review of psychology, 55(1):591–621, 2004

2004
[14]

Coppock and D

A. Coppock and D. P. Green. Assessing the correspondence between experimental results obtained in the lab and field: A review of recent social science research.Political Science Research and Methods, 3:113–131, 2015

2015
[15]

Chicago Studies in American Politics

Alexander Coppock.Persuasion in parallel : how information changes minds about politics. Chicago Studies in American Politics. The University of Chicago Press, Chicago, 1 edition, 2022

2022
[16]

When treatments are tweets: A network mobilization experiment over twitter.Political behavior, 38(1):105–128, 2016

Alexander Coppock, Andrew Guess, and John Ternovski. When treatments are tweets: A network mobilization experiment over twitter.Political behavior, 38(1):105–128, 2016

2016
[17]

Schaffner, and Alicia Prevost

Mia Costa, Brian F. Schaffner, and Alicia Prevost. Walking the walk? experiments on the effect of pledging to vote on youth turnout.PloS one, 13(5):e0197066–, 2018

2018
[18]

T. H. Costello, G. Pennycook, and D. G. Rand. Durably reducing conspiracy beliefs through dialogues with ai.Science, 385:eadq1814, 2024

2024
[19]

Cryder, George Loewenstein, and Richard Scheines

Cynthia E. Cryder, George Loewenstein, and Richard Scheines. The donor is in the details.Organizational behavior and human decision processes, 120(1):15–23, 2013

2013
[20]

DellaVigna and D

S. DellaVigna and D. Pope. What motivates effort? evidence and expert forecasts.Review of Economic Studies, 85:1029–1069, 2018

2018
[21]

A Mechanism-Based Approach to Mitigating Harms from Persuasive Generative AI.arXiv preprint arXiv:2404.15058, 2024

S. El-Sayed, C. Akbulut, A. McCroskery, G. Keeling, Z. Kenton, Z. Jalan, N. Marchal, A. Manzini, T. Shevlane, S. Vallor, D. Susser, M. Franklin, S. Bridgers, H. Law, M. Rahtz, M. Shanahan, M. H. Tessler, A. Douillard, T. Everitt, and S. Brown. A mechanism-based approach to mitigating harms from persuasive generative ai, 2024. arXiv:2404.15058 [Preprint]

work page arXiv 2024
[22]

Compliance without pressure: The foot-in-the-door technique

Jonathan L Freedman and Scott C Fraser. Compliance without pressure: The foot-in-the-door technique. Journal of personality and social psychology, 4(2):195–202, 1966. 11

1966
[23]

A. S. Gerber, D. P. Green, and C. W. Larimer. Social pressure and voter turnout: Evidence from a large-scale field experiment.American Political Science Review, 102:33–48, 2008

2008
[24]

J. A. Goldstein, J. Chao, S. Grossman, A. Stamos, and M. Tomz. How persuasive is ai-generated propaganda?PNAS Nexus, 3:pgae034, 2024

2024
[25]

Hackenburg, B

K. Hackenburg, B. M. Tappin, P. Röttger, S. A. Hale, J. Bright, and H. Margetts. Scaling language model size yields diminishing returns for single-message political persuasion.Proceedings of the National Academy of Sciences, 122:e2413443122, 2025

2025
[26]

The levers of political persuasion with conversational artificial intelligence.Science, 390(6777):eaea3884, 2025

Kobi Hackenburg, Ben M Tappin, Luke Hewitt, Ed Saunders, Sid Black, Hause Lin, Catherine Fist, Helen Margetts, David G Rand, and Christopher Summerfield. The levers of political persuasion with conversational artificial intelligence.Science, 390(6777):eaea3884, 2025

2025
[27]

Hainmueller, D

J. Hainmueller, D. Hangartner, and T. Yamamoto. Validating vignette and conjoint survey experiments against real-world behavior.Proceedings of the National Academy of Sciences, 112:2395–2400, 2015

2015
[28]

Distinguishing between most important problems and issues? Public opinion quarterly, 75(3):545–555, 2011

Will Jennings and Christopher Wlezien. Distinguishing between most important problems and issues? Public opinion quarterly, 75(3):545–555, 2011

2011
[29]

F. Luciano. Hypersuasion – on ai’s persuasive power and how to deal with it.Philosophy & Technology, 37:64, 2024

2024
[30]

The influence of affective and cognitive appeals on persuasion outcomes: a cross-cultural meta-analysis.Journal of communication, 75(2):101–111, 2025

Wei Jie Reiner Ng, Ya Hui Michelle See, and Mike W L Cheung. The influence of affective and cognitive appeals on persuasion outcomes: a cross-cultural meta-analysis.Journal of communication, 75(2):101–111, 2025

2025
[31]

D. J. O’Keefe. Persuasive message pretesting using non-behavioral outcomes: Differences in attitudinal and intention effects as diagnostic of differences in behavioral effects.Journal of Communication, 71:623–645, 2021

2021
[32]

Pagano and Yuen J

Sabrina J. Pagano and Yuen J. Huo. The role of moral emotions in predicting support for political actions in post-war iraq.Political psychology, 28(2):227–255, 2007

2007
[33]

The social costs of public political participation: Evidence from a petition experiment in lebanon.The Journal of Politics, 80:1405 – 1410, 2018

Laura Paler, Leslie Marshall, and Sami Atallah. The social costs of public political participation: Evidence from a petition experiment in lebanon.The Journal of Politics, 80:1405 – 1410, 2018

2018
[34]

Data quality of platforms and panels for online behavioral research.Behavior research methods, 54(4):1643–1662, 2022

Eyal Peer, David Rothschild, Andrew Gordon, Zak Evernden, and Ekaterina Damer. Data quality of platforms and panels for online behavioral research.Behavior research methods, 54(4):1643–1662, 2022

2022
[35]

Springer Series in Social Psychology

Richard E Petty and John T Cacioppo.Communication and Persuasion: Central and Peripheral Routes to Attitude Change. Springer Series in Social Psychology. Springer New York, New York, NY, 1 edition, 1986

1986
[36]

Saccardo, H

S. Saccardo, H. Dai, M. A. Han, S. Vangala, J. Hoo, and J. Fujimoto. Field testing the transferability of behavioural science knowledge on promoting vaccinations.Nature Human Behaviour, 8:878–890, 2024

2024
[37]

Anticipated regret as an additional predictor in the theory of planned behaviour: A meta-analysis.British journal of social psychology, 47(4):589–606, 2008

Tracy Sandberg and Mark Conner. Anticipated regret as an additional predictor in the theory of planned behaviour: A meta-analysis.British journal of social psychology, 47(4):589–606, 2008

2008
[38]

Large language models are more persuasive than incentivized human persuaders.arXiv preprint arXiv:2505.09662,

P. Schoenegger, F. Salvi, J. Liu, X. Nan, R. Debnath, B. Fasolo, E. Leivada, G. Recchia, F. Günther, A. Zarifhonarvar, J. Kwon, Z. U. Islam, M. Dehnert, D. Y. H. Lee, M. G. Reinecke, D. G. Kamper, M. Kobaş, A. Sandford, J. Kgomo, L. Hewitt, S. Kapoor, K. Oktar, E. E. Kucuk, B. Feng, C. R. Jones, I. Gainsburg, S. Olschewski, N. Heinzelmann, F. Cruz, B. M. ...

work page arXiv 2025
[39]

Gollwitzer

Paschal Sheeran, Olivia Listrom, and Peter M. Gollwitzer. The when and how of planning: Meta-analysis of the scope and components of implementation intentions in 642 tests.European review of social psychology, 36(1):162–194, 2025

2025
[40]

Paschal Sheeran and Thomas L. Webb. The intention-behavior gap.Social and personality psychology compass, 10(9):503–518, 2016

2016
[41]

Simchon, M

A. Simchon, M. Edwards, and S. Lewandowsky. The persuasive effects of political microtargeting in the age of generative artificial intelligence.PNAS Nexus, 3:pgae035, 2024

2024
[42]

Representativeness and response validity across nine opt-in online samples, Jul 2025

Michael N Stagnaro, James Druckman, Adam J Berinsky, Antonio A Arechar, Robb Willer, and David G Rand. Representativeness and response validity across nine opt-in online samples, Jul 2025

2025
[43]

The impact of advanced ai systems on democracy.Nature Human Behaviour, pages 1–11, 2025

Christopher Summerfield, Lisa P Argyle, Michiel Bakker, Teddy Collins, Esin Durmus, Tyna Eloundou, Iason Gabriel, Deep Ganguli, Kobi Hackenburg, Gillian K Hadfield, et al. The impact of advanced ai systems on democracy.Nature Human Behaviour, pages 1–11, 2025. 12

2025
[44]

Ben M. Tappin. For digital mass persuasion, exposure matters more than persuasiveness, December
[45]

Martijn van Zomeren, Tom Postmes, and Russell Spears. Toward an integrative social identity model of collective action: A quantitative research synthesis of three socio-psychological perspectives.Psychological bulletin, 134(4):504–535, 2008. 13

2008