Durably reducing conspiracy beliefs through dialogues with AI
The DOI matches its Crossref record; 101 of the 101 references it cites were found in a registry or an index.
Abstract
Conspiracy theory beliefs are notoriously persistent. Influential hypotheses propose that they fulfill important psychological needs, thus resisting counterevidence. Yet previous failures in correcting conspiracy beliefs may be due to counterevidence being insufficiently compelling and tailored. To evaluate this possibility, we leveraged developments in generative artificial intelligence and engaged 2190 conspiracy believers in personalized evidence-based dialogues with GPT-4 Turbo. The intervention reduced conspiracy belief by ~20%. The effect remained 2 months later, generalized across a wide range of conspiracy theories, and occurred even among participants with deeply entrenched beliefs. Although the dialogues focused on a single conspiracy, they nonetheless diminished belief in unrelated conspiracies and shifted conspiracy-related behavioral intentions. These findings suggest that many conspiracy theory believers can revise their views if presented with sufficiently compelling evidence.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
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?
In the citing paper
“This supports the view that the power of AI debunking comes from the facts and evidence it provides (3, 4), rather than any special characteristics of, or attitudes towards AI or resistance to human-delivered information.”
From Dialogues with large language models reduce conspiracy beliefs even when the AI is perceived as human (2025) · cites this paper as [3]
In the citing paper
“Recent experiments have also shown that conversational generative AI fact-checking can yield positive outcomes (e.g., Costello et al. 2024; Lu 2025).”
From Generative AI and misinformation: a scoping review of the role of generative AI in the generation, detection, mitigation, and impact of misinformation (2026) · cites this paper as Costello et al. 2024
In the citing paper
“For example, while LLMs’ persuasive writing has been associated with deception (25, 52) and harmful outcomes (27), recent studies have shown that it can equally be used to dissuade users of their beliefs in conspiracy theories (153), or drive users’ willingness to make donations for positive outcomes (55).”
From The benefits and dangers of anthropomorphic conversational agents (2025) · cites this paper as [153]
In the citing paper
“Critically, however, AI-powered debunking experiments have thus far involved more than just facts—they also rely on messages expressly delivered by a large language model (LLM).”
From Dialogues with large language models reduce conspiracy beliefs even when the AI is perceived as human (2025) · cites this paper as [3]
In the citing paper
“Although conspiracy beliefs are often viewed as resistant to correction, recent evidence shows that personalized, fact-based dialogues with a large language model (LLM) can reduce them.”
From Dialogues with large language models reduce conspiracy beliefs even when the AI is perceived as human (2025) · cites this paper as [3]
2 more claims about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
(1996) · Cited by 20,179
Petty, R. E., & Cacioppo, J. T. (1986) · doi:10.1007/978-1-4612-4964-1 · Cited by 5,219
Pennycook, G., & Rand, D. G. (2019) · Cognition · doi:10.1016/j.cognition.2018.06.011 · Cited by 2,022
Self-citationProceedings of the 2020 CHI Conference on Human Factors in Computing Systems
(2020) · doi:10.1145/3313831 · Cited by 1,951
Athey, S., Tibshirani, J., & Wager, S. (2019) · The Annals of Statistics · doi:10.1214/18-aos1709 · Cited by 1,870
Recent Advances in Natural Language Processing via Large Pre-Trained Language Models: A Survey
Min, B., Ross, H., Sulem, E., Veyseh, A. P. B., Nguyen, T. H., Sainz, O., Agirre, E., Heinz, I., & Roth, D. (2021) · arXiv · doi:10.48550/arxiv.2111.01243 · Cited by 1,387
Recent Advances in Natural Language Processing via Large Pre-trained Language Models: A Survey
Min, B., Ross, H., Sulem, E., Veyseh, A. P. B., Nguyen, T. H., Sainz, O., Agirre, E., Heintz, I., & Roth, D. (2024) · ACM Computing Surveys · doi:10.1145/3605943 · Cited by 1,387
The Psychology of Conspiracy Theories
Douglas, K. M., Sutton, R. M., & Cichocka, A. (2017) · Current Directions in Psychological Science · doi:10.1177/0963721417718261 · Cited by 1,316 · cited 3×
“Popular explanations propose that people adopt conspiracy theories to sate underlying psychic “ needs ” or motivations, such as thedesirefor control over one ’ s environment and experiences ( 15 ), certainty and predictability ( 20 ), security and stability ( 21 ), and uniqueness ( 22 ).”
ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks
Gilardi, F., Alizadeh, M., & Kubli, M. (2023) · arXiv · doi:10.48550/arxiv.2303.15056 · Cited by 1,207
ChatGPT outperforms crowd workers for text-annotation tasks
Gilardi, F., Alizadeh, M., & Kubli, M. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2305016120 · Cited by 1,197
Research works printed but not linked: 2
- M. G. Napolitano “Conspiracy theories and resistance to evidence ” thesis University of California Irvine (2022).
- R. Hofstadter, The Paranoid Style in American Politics (Knopf Doubleday Publishing Group, 1964).
Cited by
Items 301–324 of 324
Reconstruction of interpretable causal network dynamics from time series data
Gao, X., Sun, X., Wu, T., An, S., An, F., Leng, S., Wei, H., Zhang, Y., Marwan, N., & Kurths, J. (2026) · Physics Reports · doi:10.1016/j.physrep.2026.07.004 · Cited by 2
On the epistemic structure of voluntary servitude in the infosphere
Rodrigues, T. V. (2026) · AI & Society · doi:10.1007/s00146-026-03342-w
Toward an integrated and globally equitable understanding of political polarization
Falkenberg, M., Cinelli, M., Galeazzi, A., Bail, C. A., Benito, R. M., Bruns, A., Gruzd, A. A., Lazer, D., Lee, J. K., McCoy, J., et al. (2026) · Nature Communications · doi:10.1038/s41467-026-76110-5 · Cited by 15
Self-citationCONSPIR ED: A Dataset for Cognitive Traits of Conspiracy Theories and Large Language Model Safety
Bates, L., Glockner, M., Nakov, P., & Gurevych, I. (2026) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl.a.794 · Cited by 1
Beyond Text: Embodiment and Robots
Perry, A., & Cameron, C. D. (2026) · Cambridge University Press eBooks · doi:10.1017/9781009639057.016
From Social Media to Empathic Artificial Intelligence: Applying Past Lessons to Future Technologies
Rodriguez, M., Motyl, M., & Schroeder, J. (2026) · Cambridge University Press eBooks · doi:10.1017/9781009639057.023
How AI amplifies, mitigates, and creates social biases: A tripartite framework
Gai, P. J. (2026) · Current Opinion in Psychology · doi:10.1016/j.copsyc.2026.102426
Makhortykh, M., & Kuznetsova, E. (2026) · Digital Hate Review · doi:10.66144/xgilx5h
Human preferences are susceptible to covertly misaligned AI advice
Sabour, S., Liu, J. M., Liu, S., Yao, C. Z., Cui, S., Zhang, W., Zhang, X., Cao, Y., Bhat, A., Guan, J., et al. (2026) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2600684123 · Cited by 4
Nitchals, C. (2026) · AI and Ethics · doi:10.1007/s43681-026-01379-7 · Cited by 1
“Controlled experiments demonstrate that personal - ized dialogue with a capable LLM can change judgments and beliefs, including political attitudes and conspiracy beliefs [ 16 – 18 ].”
InfluentialMeatless minds vs. beast machines: Are simulated subjects impossible?
Dołęga, K., & Cleeremans, A. (2026) · Behavioral and Brain Sciences · doi:10.1017/s0140525x25103889
Corpus-level behavioral intensity in human–AI conversations
Mera-Navarrete, A., Revelo-Tapia, S., Beltrán-Morales, J., & Villegas-Ch, W. E. (2026) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2026.1914081
“Similarly, dialogues with AI have demonstrated the capacity to alter persistent beliefs in sensitive contexts, such as conspiracy theories (Costello et al., 2024), while education, health, and scientific writing document operational benefits together with risks of uncritical use, hallucinations, and…”
When AI talks politics: How reliable is ChatGPT?
Hartland, A., Braun, D., Carteny, G., & Navarrete, R. M. (2026) · Electoral Studies · doi:10.1016/j.electstud.2026.103165
LENS—LLM Effects on News Scrutiny: Influence, Delegation, and Memory in AI-Mediated News
Reina, L., Rocca, S. L., Longoni, F., & Riva, G. (2026) · Cyberpsychology Behavior and Social Networking · doi:10.1177/21522715261492446
Zhu, Z., & Li, Y. (2026) · Telematics and Informatics · doi:10.1016/j.tele.2026.102464 · Cited by 1
How Model Choice and Memory Shape Preference Consistency in Large Language Models
Chung, Y. (2026) · Sociological Methods & Research · doi:10.1177/00491241261492203
How Artificial Intelligence Shapes Human Narratives
Priniski, J. H. (2026) · Journal of Language and Social Psychology · doi:10.1177/0261927x261493082
A beta way: A tutorial on Bayesian beta regression for psychological research
Geller, J., Kubinec, R., Parlett-Pelleriti, C., & Vuorre, M. (2026) · Behavior Research Methods · doi:10.3758/s13428-026-03106-w
“These include the proportion of correct responses on a test (e.g., Kornell and Bjork (2008)), the proportion of time a participant fixates on a stimulus in an eye-tracking task (e.g., James et al., 2025), or the proportion of respondents endorsing a particular belief (e.g., Costello et al., 2024).”
Liu, N. (2026) · Companion Publication of the International Conference on Multimodal Interaction (ICMI Companion) · doi:10.1145/3776591.3832500
A perfect storm: How social media may heighten older adults’ vulnerability to health misinformation
Mennie, K., LaCour, M., & Morse, M. (2026) · Educational Gerontology · doi:10.1080/03601277.2026.2744721
Lee, J., Kim, S., Lee, H., & Hwang, J. (2026) · International Journal of Human-Computer Interaction · doi:10.1080/10447318.2026.2740422
Mitigating Hallucination in AI Using a Novel Approach
Gayathri, R. M., Kanmani, P., & Vijayan, A. (2026) · Lecture notes in electrical engineering · doi:10.1007/978-981-95-7449-0_21
Haque, M. N. (2025) · Research journal in business and economics. · doi:10.61424/rjbe.v3i4.1082
Attractor States Emerge in Multi-Turn LLM Conversations
Ko, T.-W., & Geiping, J. (2026) · arXiv · doi:10.48550/arxiv.2606.30571
“Taubenfeld et al. (2024); Costello et al. (2024) further demonstrate that LLM agents in simulated debates tend to conform to the model’s inherent social biases regardless of assigned perspective.”
Metadata
- Authors
- Costello, T. H., Pennycook, G., & Rand, D. G.
- Affiliations
- American University (US)
- Cornell University (US)
- Massachusetts Institute of Technology (US)
- Corresponding author
- Costello, Thomas H.
- Venue
- Science
- Published
- 2024-09-13
- Volume
- 385
- Issue
- 6714
- Pages
- eadq1814-eadq1814
- ISSN
- 0036-8075 (ISSN-L)
- 1095-9203
- OpenAlex
- W4402474910
- Cited by
- 354
- References
- 79
- Citations by year
- 2026: 175 · 2025: 162 · 2024: 16
- FWCI
- 252.37
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- Science (publishedVersion)
- PubMed (publishedVersion)
- Topics
- Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
- Social Sciences › Psychology › Social Psychology › Paranormal Experiences and Beliefs
- Social Sciences › Sociology and Political Science › Social and Intergroup Psychology
- Keywords
- conspiracy beliefs, conspiracy theories, belief revision, counterevidence, personalized dialogue, generative AI, GPT-4 Turbo, belief persistence, behavioral intention
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.
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
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.