pith. sign in

arxiv: 2605.23890 · v1 · pith:F4UR6QWCnew · submitted 2026-05-22 · 💻 cs.CY · cs.HC

Divergent Paths to Depolarization: Dialogue Design Determines the Prosocial Benefits of AI-Assisted Political Argumentation

Jianlong Zhu , Syed Muhammad Jhon Raza Naqvi , Carolin-Theresa Ziemer , Usman Naseem , Ingmar Weber This is my paper

Pith reviewed 2026-05-25 02:40 UTC · model grok-4.3

classification 💻 cs.CY cs.HC
keywords AI dialoguepolitical polarizationattitude congruencecognitive empathydepolarizationhuman-AI interactiononline experiment
0
0 comments X

The pith

AI dialogues that match a user's political views reduce polarization faster than those that challenge them.

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

Through a two-session experiment involving 469 participants, the study demonstrates that AI chatbots can facilitate political argumentation with varying effects based on whether the dialogue supports or opposes the user's stance. Attitude-congruent dialogues lead to more substantial short-term decreases in both affective and opinion polarization. Attitude-incongruent dialogues, while less effective immediately on empathy, show increases in cognitive trait empathy over the following two weeks. This matters for designing scalable tools that help people engage with opposing political views without requiring in-person interactions.

Core claim

The results reveal that attitude-congruent dialogues with an AI produce greater immediate reductions in affective and opinion polarization compared to attitude-incongruent dialogues. Attitude-incongruent dialogues yield weaker cognitive state empathy than a non-AI task but increase cognitive trait empathy across the two-week interval, pointing to time-dependent effects from actively constructing counter-attitudinal arguments.

What carries the argument

Attitude congruence versus incongruence in the design of AI-assisted political argumentation dialogues, which drives distinct patterns of polarization reduction and empathy development.

If this is right

  • Congruent AI dialogues can serve as an effective method for achieving rapid reductions in political polarization.
  • Incongruent dialogues may be suited for interventions aimed at long-term empathy building.
  • The choice of dialogue format in AI systems determines the timing and nature of prosocial benefits.
  • AI-mediated exercises offer a structured alternative for citizens to practice open-mindedness on contested issues.
  • Benefits from incongruent arguments may accumulate gradually rather than appearing instantly.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • AI platforms could implement dynamic dialogue modes that shift from congruent to incongruent based on user goals or session number.
  • These findings could inform applications in reducing divides on topics like climate policy or social justice beyond the tested political issues.
  • Combining AI dialogues with real-world follow-ups might enhance the observed empathy gains.
  • Examining whether similar patterns hold in group settings or with multiple AI agents would test the robustness of the design principle.

Load-bearing premise

The premise that self-reported changes in polarization and empathy measures accurately reflect genuine psychological shifts attributable to the dialogue conditions rather than other influences over two weeks.

What would settle it

Observing no differences in polarization or empathy between congruent and incongruent conditions in a study that uses behavioral tasks, such as actual intergroup contact or donation decisions, and tightly controls for external events would challenge the paper's findings.

read the original abstract

Argumentative dialogues across political divides can reduce polarization, yet opportunities for citizens to engage with opposing views in accessible and structured ways remain limited. AI dialogue partners offer a scalable framework for such open-mindedness exercises, but how the format of human-AI dialogues shapes their benefits remains unclear. In a two-session online experiment, 469 US participants were assigned to argue either for or against their own attitude on a contested political issue with an AI chatbot. Our experimental findings show attitude-congruent dialogues produced greater immediate reduction in both affective and opinion polarization than attitude-incongruent dialogues. By contrast, attitude-incongruent dialogues elicited weaker cognitive state empathy than the non-AI reference task but increased cognitive trait empathy in the two-week period between sessions, suggesting the effects of active generation of attitude-incongruent arguments may emerge over time. These findings highlight dialogue design as a key determinant of effective AI-mediated behavioral interventions.

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.

Referee Report

2 major / 2 minor

Summary. The paper reports results from a two-session online experiment (N=469 US participants) in which individuals were randomly assigned to engage in attitude-congruent or attitude-incongruent argumentative dialogues with an AI chatbot on a contested political issue. It claims that congruent dialogues produced larger immediate reductions in affective and opinion polarization than incongruent dialogues, while incongruent dialogues produced weaker immediate cognitive state empathy than a non-AI reference task but larger increases in cognitive trait empathy over the two-week interval between sessions.

Significance. If the results hold after addressing the noted limitations, the work would demonstrate that dialogue format is a critical design variable for AI-mediated depolarization interventions, with short-term polarization effects favoring congruent arguments and delayed empathy effects favoring incongruent arguments. The randomized assignment and two-session structure are strengths that allow causal contrasts between conditions.

major comments (2)
  1. [Results (two-week analysis)] Results section (two-week follow-up): The reported increase in cognitive trait empathy for the attitude-incongruent condition is presented as an effect of the dialogue without any reported controls or measures for intervening events, media consumption, or additional discussions during the two-week gap; this leaves open the possibility that external factors rather than the assigned condition drive the trait change, directly weakening the central claim that incongruent dialogues produce delayed prosocial benefits.
  2. [Methods and Results] Methods and Results: No information is provided on statistical tests, effect sizes, confidence intervals, attrition rates, or pre-registered analysis plans for the polarization and empathy deltas; without these, it is impossible to assess whether the directional findings survive correction for multiple comparisons or are robust to participant dropout.
minor comments (2)
  1. [Abstract] Abstract and introduction: The non-AI reference task is mentioned but never described in sufficient detail to allow readers to interpret the state-empathy comparison.
  2. [Methods] The manuscript would benefit from explicit discussion of the convergent validity of the self-report polarization and empathy scales used, including any behavioral correlates collected.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their detailed and constructive comments. We respond to each major point below and will revise the manuscript to strengthen the presentation of limitations and statistical details.

read point-by-point responses

  1. Referee: [Results (two-week analysis)] Results section (two-week follow-up): The reported increase in cognitive trait empathy for the attitude-incongruent condition is presented as an effect of the dialogue without any reported controls or measures for intervening events, media consumption, or additional discussions during the two-week gap; this leaves open the possibility that external factors rather than the assigned condition drive the trait change, directly weakening the central claim that incongruent dialogues produce delayed prosocial benefits.

    Authors: We agree this is a substantive limitation. The design relies on random assignment to isolate condition effects, but without measures of intervening events we cannot rule out external influences on the two-week trait empathy change. The manuscript already describes the finding as suggestive rather than definitive. We will revise the Discussion to explicitly list the absence of intervening-event controls as a key limitation and adjust language to avoid implying strong causal attribution for the delayed effect. revision: yes

  2. Referee: [Methods and Results] Methods and Results: No information is provided on statistical tests, effect sizes, confidence intervals, attrition rates, or pre-registered analysis plans for the polarization and empathy deltas; without these, it is impossible to assess whether the directional findings survive correction for multiple comparisons or are robust to participant dropout.

    Authors: We will add a dedicated Statistical Analysis subsection in Methods that reports all tests (including exact procedures for deltas), effect sizes, confidence intervals, attrition rates from initial N to follow-up, and handling of multiple comparisons. We will also note that the analysis plan was not pre-registered and discuss this as a limitation. These additions will allow readers to evaluate robustness directly. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical experiment with direct observations

full rationale

The paper describes a two-session online experiment with 469 participants assigned to dialogue conditions, reporting pre/post changes in self-reported polarization and empathy measures. No equations, derivations, fitted parameters, or model-based predictions appear in the abstract or described methods. All results are presented as statistical comparisons of observed data across conditions rather than quantities defined in terms of prior outputs or self-citations. The study is self-contained against external benchmarks with no load-bearing self-referential steps.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard social-science assumptions about the validity of self-report measures and the ability to isolate dialogue-format effects in an online setting; no mathematical free parameters or invented entities are introduced.

axioms (2)
  • domain assumption Self-reported attitudes, polarization, and empathy scales accurately reflect participants' internal psychological states.
    All outcome measures are questionnaire-based and the abstract treats changes in these scores as direct evidence of the claimed effects.
  • domain assumption The two-week interval contains no systematic external influences that differentially affect the congruent and incongruent groups.
    The delayed trait-empathy increase is attributed to the initial dialogue condition.

pith-pipeline@v0.9.0 · 5715 in / 1412 out tokens · 33198 ms · 2026-05-25T02:40:02.435417+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

62 extracted references · 62 canonical work pages

  1. [1]

    & Lelkes, Y

    Iyengar, S., Sood, G. & Lelkes, Y. Affect, Not Ideology.Public Opinion Quarterly 76, 405–431 (2012). URL https://academic.oup.com/poq/article-lookup/doi/10 .1093/poq/nfs038

    work page 2012

  2. [2]

    H., Ankori-Karlinsky, L.-O., Levy-Vene, A

    Pasek, M. H., Ankori-Karlinsky, L.-O., Levy-Vene, A. & Moore-Berg, S. L. Misperceptions about out-partisans’ democratic values may erode democracy.Sci- entific Reports12, 16284 (2022). URL https://www.nature.com/articles/s41598 -022-19616-4

    work page 2022

  3. [3]

    C.Hearing the Other Side: Deliberative versus Participatory Democracy 1 edn (Cambridge University Press, 2006)

    Mutz, D. C.Hearing the Other Side: Deliberative versus Participatory Democracy 1 edn (Cambridge University Press, 2006). URL https://www.cambridge.org/co re/product/identifier/9780511617201/type/book

    work page arXiv 2006

  4. [4]

    Schneider, F. M. & Weinmann, C. In Need of the Devil’s Advocate? The Impact of Cross-Cutting Exposure on Political Discussion.Political Behavior45, 373–394 (2023). URL https://doi.org/10.1007/s11109-021-09706-w

    work page doi:10.1007/s11109-021-09706-w 2023

  5. [5]

    & Spruyt, B

    Caluwaerts, D., Bernaerts, K., Kesberg, R., Smets, L. & Spruyt, B. Deliberation and polarization: a multi-disciplinary review.Frontiers in Political Science5, 14 1127372 (2023). URL https://www.frontiersin.org/articles/10.3389/fpos.2023.1 127372/full

    work page doi:10.3389/fpos.2023.1 2023

  6. [6]

    America in One Room

    Fishkin, J., Siu, A., Diamond, L. & Bradburn, N. Is Deliberation an Antidote to Extreme Partisan Polarization? Reflections on “America in One Room”.Amer- ican Political Science Review115, 1464–1481 (2021). URL https://www.camb ridge.org/core/journals/american-political-science-review/article/is-deliberatio n-an-antidote-to-extreme-partisan-polarization-re...

    work page 2021

  7. [7]

    Kennedy, R. R. The power of in-class debates.Active Learning in Higher Education10, 225–236 (2009). URL https://doi.org/10.1177/1469787409343186

    work page doi:10.1177/1469787409343186 2009

  8. [8]

    Chen, X., Wang, L., Zhai, X. & Li, Y. Exploring the Effects of Argument Map-Supported Online Group Debate Activities on College Students’ Critical Thinking.Frontiers in Psychology13(2022). URL https://www.frontiersin.or g/journals/psychology/articles/10.3389/fpsyg.2022.856462/full

    work page doi:10.3389/fpsyg.2022.856462/full 2022

  9. [9]

    The Impact of Competitive Debate on Managing the Conflict Communication Strategies of Italian Students.Argumentation and Advocacy51, 123–131 (2014)

    De Conti, M. The Impact of Competitive Debate on Managing the Conflict Communication Strategies of Italian Students.Argumentation and Advocacy51, 123–131 (2014). URL https://doi.org/10.1080/00028533.2014.11821843

    work page doi:10.1080/00028533.2014.11821843 2014

  10. [10]

    Budesheim, T. L. & Lundquist, A. R. Consider the Opposite: Opening Minds Through In-Class Debates on Course-Related Controversies.Teaching of Psy- chology(1999). URL https://www.tandfonline.com/doi/abs/10.1207/s1532802 3top2602 5

    work page doi:10.1207/s1532802 1999

  11. [11]

    Debate, Education, and Polarization: Why Polarization is not a Debate Side Effect.Cogency : journal of reasoning and argumentation5(2013)

    De Conti, M. Debate, Education, and Polarization: Why Polarization is not a Debate Side Effect.Cogency : journal of reasoning and argumentation5(2013)

    work page 2013

  12. [12]

    S., Huber, G

    Gerber, A. S., Huber, G. A., Doherty, D. & Dowling, C. M. Disagreement and the Avoidance of Political Discussion: Aggregate Relationships and Differences across Personality Traits.American Journal of Political Science56, 849–874 (2012). URL https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5907.2011.00571.x

    work page doi:10.1111/j.1540-5907.2011.00571.x 2012

  13. [13]

    E., Aubin, C

    Eddy, K., Lipka, M., Matsa, K. E., Aubin, C. S. & Shearer, E. Why Americans discuss the news – or don’t – with others (2026). URL https://www.pewresearc h.org/journalism/2026/02/11/why-americans-discuss-the-news-or-dont-with-o thers/. Section: News Habits & Media

    work page 2026

  14. [14]

    Boland, F. K. & Davidai, S. Zero-sum beliefs and the avoidance of political conversations.Communications Psychology2, 43 (2024). URL https://www.na ture.com/articles/s44271-024-00095-4

    work page 2024

  15. [15]

    The Online Disinhibition Effect.CyberPsychology & Behavior7, 321–326 (2004)

    Suler, J. The Online Disinhibition Effect.CyberPsychology & Behavior7, 321–326 (2004). URL https://journals.sagepub.com/doi/full/10.1089/1094931041291295. 15

    work page doi:10.1089/1094931041291295 2004

  16. [16]

    URL https://www.nature.com/articles/s41562-023-01655-0

    Combs, A.et al.Reducing political polarization in the United States with a mobile chat platform.Nature Human Behaviour7, 1454–1461 (2023). URL https://www.nature.com/articles/s41562-023-01655-0

    work page 2023

  17. [17]

    De Jong, J. F. Cross-partisan discussions reduced political polarization between UK voters, but less so when they disagreed.Communications Psychology2, 5 (2024). URL https://www.nature.com/articles/s44271-023-00051-8

    work page 2024

  18. [18]

    Croes, E. A. J., Antheunis, M. L., Van Der Lee, C. & De Wit, J. M. S. Digital Confessions: The Willingness to Disclose Intimate Information to a Chatbot and its Impact on Emotional Well-Being.Interacting with Computers36, 279–292 (2024). URL https://academic.oup.com/iwc/article/36/5/279/7692197

    work page 2024

  19. [19]

    Lu, L., Tormala, Z. L. & Duhachek, A. How AI sources can increase openness to opposing views.Scientific Reports15, 17170 (2025). URL https://www.nature .com/articles/s41598-025-00791-z

    work page 2025

  20. [20]

    & Tarle, P

    Willis, M., Meacham, B. & Tarle, P. Bridging the Divide: Using LLMs to Reshape Difficult Disagreements.Philosophy & Technology39, 78 (2026). URL https: //link.springer.com/10.1007/s13347-026-01076-7

    work page doi:10.1007/s13347-026-01076-7 2026

  21. [21]

    Hruschka, T. M. J. & Appel, M. Reducing political polarization through conversa- tions with artificial intelligence.Journal of Computer-Mediated Communication 31, zmag003 (2026). URL https://academic.oup.com/jcmc/article/doi/10.1093/ jcmc/zmag003/8551324

    work page 2026

  22. [22]

    Costello, Gordon Pennycook, and David G

    Costello, T. H., Pennycook, G. & Rand, D. G. Durably reducing conspiracy beliefs through dialogues with AI.Science385, eadq1814 (2024). URL https: //www.science.org/doi/10.1126/science.adq1814

    work page doi:10.1126/science.adq1814 2024

  23. [23]

    & Mekik, C

    Sevi, S. & Mekik, C. S. AI Weakens, But Does Not Strengthen, Political Attitudes (2026). URL https://osf.io/ukjnc v1

    work page 2026

  24. [24]

    & Zhang, Y

    Farag, Y.et al.Goldberg, Y., Kozareva, Z. & Zhang, Y. (eds)Opening up Minds with Argumentative Dialogues. (eds Goldberg, Y., Kozareva, Z. & Zhang, Y.)Findings of the Association for Computational Linguistics: EMNLP 2022, 4569–4582 (Association for Computational Linguistics, Abu Dhabi, United Arab Emirates, 2022). URL https://aclanthology.org/2022.findings...

    work page 2022

  25. [25]

    & Cardie, C

    Durmus, E. & Cardie, C. Exploring the Role of Prior Beliefs for Argument Persuasion 1035–1045 (2018). URL http://aclweb.org/anthology/N18-1094

    work page 2018

  26. [26]

    URL https://www.nature.com/articles/s41586-021-03215-w

    Slonim, N.et al.An autonomous debating system.Nature591, 379–384 (2021). URL https://www.nature.com/articles/s41586-021-03215-w

    work page 2021

  27. [27]

    Using AI Persuasion to Reduce Political Polarization (2025)

    Walter, J. Using AI Persuasion to Reduce Political Polarization (2025). URL https://www.zew.de/publikationen/using-ai-persuasion-to-reduce-political-pol 16 arization

    work page 2025

  28. [28]

    Miller, R. L. & Wozniak, W. Counter-attitudinal advocacy: Effort vs. self- generation of arguments.Current Research in Social Psychology6, 46–55 (2001)

    work page 2001

  29. [29]

    & Wegener, D

    Xu, M. & Wegener, D. T. Persuasive Benefits of Self-Generated Arguments: Moderation and Mechanism.Social Psychological and Personality Science15, 60–69 (2024). URL https://doi.org/10.1177/19485506221146612

    work page doi:10.1177/19485506221146612 2024

  30. [30]

    E., Wheeler, S

    Petty, R. E., Wheeler, S. C. & Tormala, Z. L. inPersuasion and Attitude Change 1 edn, (ed.Weiner, I. B.)Handbook of Psychology353–382 (Wiley, 2003). URL https://onlinelibrary.wiley.com/doi/10.1002/0471264385.wei0515

    work page doi:10.1002/0471264385.wei0515 2003

  31. [31]

    Gurin, P., Nagda, B. R. A. & Z´ u˜ niga, X. (eds)Dialogue Across Difference: Prac- tice, Theory, and Research on Intergroup Dialogue(Russell Sage Foundation, 2013). URL https://www.jstor.org/stable/10.7758/9781610448055

    work page doi:10.7758/9781610448055 2013

  32. [32]

    Do Ideologically Intolerant People Benefit From Intergroup Contact? Current Directions in Psychological Science20, 154–159 (2011)

    Hodson, G. Do Ideologically Intolerant People Benefit From Intergroup Contact? Current Directions in Psychological Science20, 154–159 (2011). URL https: //journals.sagepub.com/doi/10.1177/0963721411409025

    work page doi:10.1177/0963721411409025 2011

  33. [33]

    & Felise, K

    Bliuc, A.-M., Bouguettaya, A. & Felise, K. D. Online Intergroup Polarization Across Political Fault Lines: An Integrative Review.Frontiers in Psychology12, 641215 (2021). URL https://www.frontiersin.org/articles/10.3389/fpsyg.2021. 641215/full

    work page doi:10.3389/fpsyg.2021 2021

  34. [34]

    A.et al.Exposure to opposing views on social media can increase political polarization.Proceedings of the National Academy of Sciences115, 9216–9221 (2018)

    Bail, C. A.et al.Exposure to opposing views on social media can increase political polarization.Proceedings of the National Academy of Sciences115, 9216–9221 (2018). URL https://pnas.org/doi/full/10.1073/pnas.1804840115

    work page doi:10.1073/pnas.1804840115 2018

  35. [35]

    & Lindahl, M

    Boss´ er, U. & Lindahl, M. G. Students’ Use of Open-Minded Attitude and Elabo- rate Talk in Group Discussion and Role-Playing Debate on Socioscientific Issues. Eurasia Journal of Mathematics, Science and Technology Education16, em1910 (2020). URL https://www.ejmste.com/article/students-use-of-open-minded-att itude-and-elaborate-talk-in-group-discussion-an...

    work page 2020

  36. [36]

    Open to Debate: Reducing Polarization by Approaching Political Argu- ment as Negotiation.Negotiation Journal37, 361–391 (2021)

    Iwry, J. Open to Debate: Reducing Polarization by Approaching Political Argu- ment as Negotiation.Negotiation Journal37, 361–391 (2021). URL https: //doi.org/10.1111/nejo.12370

    work page doi:10.1111/nejo.12370 2021

  37. [37]

    E., Cooper, J

    Linder, D. E., Cooper, J. & Jones, E. E. Decision freedom as a determinant of the role of incentive magnitude in attitude change.Journal of Personality and Social Psychology6, 245–254 (1967)

    work page 1967

  38. [38]

    & Rothmund, T

    Ziemer, C.-T. & Rothmund, T. Psychological underpinnings of misinformation countermeasures: A systematic scoping review.Journal of Media Psychology: 17 Theories, Methods, and Applications36, 397–409 (2024). Place: Germany

    work page 2024

  39. [39]

    Rathje, S., Roozenbeek, J., Van Bavel, J. J. & van der Linden, S. Accuracy and social motivations shape judgements of (mis)information.Nature Human Behaviour7, 892–903 (2023). URL https://www.nature.com/articles/s41562-0 23-01540-w

    work page 2023

  40. [40]

    & Martini, C

    Ronzani, P., Panizza, F., Morisseau, T., Mattavelli, S. & Martini, C. How differ- ent incentives reduce scientific misinformation online.Harvard Kennedy School Misinformation Review(2024). URL https://misinforeview.hks.harvard.edu/ar ticle/how-different-incentives-reduce-scientific-misinformation-online/

    work page 2024

  41. [41]

    & Matsumoto, K

    Murayama, K., Matsumoto, M., Izuma, K. & Matsumoto, K. Neural basis of the undermining effect of monetary reward on intrinsic motivation.Proceedings of the National Academy of Sciences107, 20911–20916 (2010). URL https: //pnas.org/doi/full/10.1073/pnas.1013305107

    work page doi:10.1073/pnas.1013305107 2010

  42. [42]

    & Dolan, P

    Vlaev, I., King, D., Darzi, A. & Dolan, P. Changing health behaviors using financial incentives: a review from behavioral economics.BMC Public Health19, 1059 (2019). URL https://bmcpublichealth.biomedcentral.com/articles/10.1186 /s12889-019-7407-8

    work page 2019

  43. [43]

    On a Scale of State Empathy During Message Processing.Western Journal of Communication74, 504–524 (2010)

    Shen, L. On a Scale of State Empathy During Message Processing.Western Journal of Communication74, 504–524 (2010). URL https://doi.org/10.1080/ 10570314.2010.512278

    work page arXiv 2010

  44. [44]

    Davis, M. H. Measuring individual differences in empathy: Evidence for a multi- dimensional approach.Journal of Personality and Social Psychology44, 113–126 (1983). URL https://doi.apa.org/doi/10.1037/0022-3514.44.1.113

    work page doi:10.1037/0022-3514.44.1.113 1983

  45. [45]

    M., V´ etois, M

    Klimecki, O. M., V´ etois, M. & Sander, D. The impact of empathy and perspective- taking instructions on proponents and opponents of immigration.Humanities and Social Sciences Communications7, 91 (2020). URL https://www.nature.c om/articles/s41599-020-00581-0

    work page 2020

  46. [46]

    G.et al.Megastudy testing 25 treatments to reduce antidemocratic attitudes and partisan animosity.Science386, eadh4764 (2024)

    Voelkel, J. G.et al.Megastudy testing 25 treatments to reduce antidemocratic attitudes and partisan animosity.Science386, eadh4764 (2024). URL https: //www.science.org/doi/10.1126/science.adh4764

    work page doi:10.1126/science.adh4764 2024

  47. [47]

    E., Lelkes, Y

    Holliday, D. E., Lelkes, Y. & Westwood, S. J. Why depolarization is hard: Evaluating attempts to decrease partisan animosity in America.Proceedings of the National Academy of Sciences122, e2508827122 (2025). URL https: //pnas.org/doi/10.1073/pnas.2508827122

    work page doi:10.1073/pnas.2508827122 2025

  48. [48]

    G., Ross, L

    Lord, C. G., Ross, L. & Lepper, M. R. Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology37, 2098–2109 (1979). URL 18 https://doi.apa.org/doi/10.1037/0022-3514.37.11.2098

    work page doi:10.1037/0022-3514.37.11.2098 2098

  49. [49]

    Warren, A. N. Navigating assigned roles for asynchronous online discussions: Examining participants’ orientation using conversation analysis.Online Learning 22(2019). URL https://olj.onlinelearningconsortium.org/index.php/olj/article /view/1512

    work page 2019

  50. [50]

    & West, R

    Salvi, F., Horta Ribeiro, M., Gallotti, R. & West, R. On the conversational persuasiveness of GPT-4.Nature Human Behaviour1–9 (2025). URL https: //www.nature.com/articles/s41562-025-02194-6

    work page 2025

  51. [51]

    P.et al.Testing theories of political persuasion using AI.Proceedings of the National Academy of Sciences122, e2412815122 (2025)

    Argyle, L. P.et al.Testing theories of political persuasion using AI.Proceedings of the National Academy of Sciences122, e2412815122 (2025). URL https: //pnas.org/doi/10.1073/pnas.2412815122

    work page doi:10.1073/pnas.2412815122 2025

  52. [52]

    URL https://www.science.org/ doi/10.1126/science.aea3884

    Hackenburg, K.et al.The levers of political persuasion with conversational arti- ficial intelligence.Science390, eaea3884 (2025). URL https://www.science.org/ doi/10.1126/science.aea3884

    work page doi:10.1126/science.aea3884 2025

  53. [53]

    A., Chao, J., Grossman, S., Stamos, A

    Goldstein, J. A., Chao, J., Grossman, S., Stamos, A. & Tomz, M. How persuasive is AI-generated propaganda?PNAS Nexus3, pgae034 (2024). URL https://ac ademic.oup.com/pnasnexus/article/doi/10.1093/pnasnexus/pgae034/7610937

    work page doi:10.1093/pnasnexus/pgae034/7610937 2024

  54. [54]

    Higgins, E. T. Self-Discrepancy Theory: What Patterns of Self-Beliefs Cause People to Suffer?22, 93–136 (1989). URL https://linkinghub.elsevier.com/retr ieve/pii/S0065260108603068

    work page 1989

  55. [55]

    & Steg, L

    Lemmen, N., Keizer, K., Bouman, T. & Steg, L. Convince Yourself to Do the Right Thing: The Effects of Provided Versus Self-Generated Arguments on Rule Compliance and Perceived Importance of Socially Desirable Behavior.Frontiers in Psychology11, 613418 (2020). URL https://www.frontiersin.org/articles/10. 3389/fpsyg.2020.613418/full

    work page arXiv 2020

  56. [56]

    S., Rothman, A

    Baldwin, A. S., Rothman, A. J., Vander Weg, M. W. & Christensen, A. J. Exam- ining causal components and a mediating process underlying self-generated health arguments for exercise and smoking cessation.Health Psychology32, 1209–1217 (2013). URL https://doi.apa.org/doi/10.1037/a0029937

    work page doi:10.1037/a0029937 2013

  57. [57]

    Bri˜ nol, P., McCaslin, M. J. & Petty, R. E. Self-generated persuasion: Effects of the target and direction of arguments.Journal of Personality and Social Psychology 102, 925–940 (2012). URL https://doi.apa.org/doi/10.1037/a0027231

    work page doi:10.1037/a0027231 2012

  58. [58]

    Kumkale, G. T. & Albarrac´ ın, D. The Sleeper Effect in Persuasion: A Meta- Analytic Review.Psychological Bulletin130, 143–172 (2004). URL https://doi. apa.org/doi/10.1037/0033-2909.130.1.143. 19

    work page doi:10.1037/0033-2909.130.1.143 2004

  59. [59]

    & Schitter, C

    Palan, S. & Schitter, C. Prolific.ac—A subject pool for online experiments.Jour- nal of Behavioral and Experimental Finance17, 22–27 (2018). URL https: //linkinghub.elsevier.com/retrieve/pii/S2214635017300989

    work page 2018

  60. [60]

    & Acquisti, A

    Peer, E., Brandimarte, L., Samat, S. & Acquisti, A. Beyond the Turk: Alternative platforms for crowdsourcing behavioral research.Journal of Experimental Social Psychology70, 153–163 (2017). URL https://linkinghub.elsevier.com/retrieve/p ii/S0022103116303201

    work page 2017

  61. [61]

    URL https: //doi.org/10.1145/3719160.3736611

    Zhu, J.et al.Learn, Explore and Reflect by Chatting: Understanding the Value of an LLM-Based Voting Advice Application Chatbot 1–19 (2025). URL https: //doi.org/10.1145/3719160.3736611

    work page doi:10.1145/3719160.3736611 2025

  62. [62]

    D., Ewell, P

    Douglas, B. D., Ewell, P. J. & Brauer, M. Data quality in online human-subjects research: Comparisons between MTurk, Prolific, CloudResearch, Qualtrics, and SONA.PLOS ONE18, e0279720 (2023). URL https://dx.plos.org/10.1371/jo urnal.pone.0279720. 20

    work page doi:10.1371/jo 2023