Generative AI in the Era of 'Alternative Facts'
Open accessAbstract
The spread of misinformation on social media platforms threatens democratic processes, contributes to massive economic losses, and endangers public health. Many efforts to address misinformation focus on a knowledge deficit model and propose interventions for improving users’ critical thinking through access to facts. Such efforts are often hampered by challenges with scalability, and by platform users’ confirmation bias. The emergence of generative AI presents promising opportunities for countering misinformation at scale across ideological barriers. In this paper, we present (1) an experiment with a simulated social media environment to measure effectiveness of misinformation interventions generated by large language models (LLMs), (2) a second experiment with personalized explanations tailored to the demographics and beliefs of users with the goal of alleviating confirmation bias, and (3) an analysis of potential harms posed by personalized generative AI when exploited for automated creation of disinformation. Our findings confirm that LLM-based interventions are highly effective at correcting user behavior (improving overall user accuracy at reliability labeling up to 47.6%). Furthermore, we find that users favor more personalized interventions when making decisions about news reliability.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
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?
Misalignment with user identity or ideology may backfire.
“Personalized corrections can improve discernment, yet misalignment with user identity or ideology may backfire (Gabriel et al. 2024), while model safeguards are neither durable nor uniformly effective (Menz et al. 2024; Węcel et al. 2023).”
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 Gabriel et al. 2024
Section 5.1 (page unavailable) reports that explanations misaligned with users’ identities received lower helpfulness ratings than aligned ones; the attributes studied include political ideology. That supports a risk from misalignment, but the paper does not show that misaligned corrections backfire by worsening discernment or other outcomes. (LLM, conf=0.84)
Experimental — accuracy not yet measuredPersonalized corrections can improve discernment.
“Personalized corrections can improve discernment, yet misalignment with user identity or ideology may backfire (Gabriel et al. 2024), while model safeguards are neither durable nor uniformly effective (Menz et al. 2024; Węcel et al. 2023).”
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 Gabriel et al. 2024
Section 5.1 (page unavailable) reports that users rated personalized explanations of veracity as more helpful when they aligned with their identities, and rated misaligned explanations as less helpful. It does not show that personalization improved users’ accuracy in discerning true from false claims, or that misalignment backfired; the accuracy gains in Section 4.2 concern nonpersonalized interventions. (LLM, conf=0.87)
Experimental — accuracy not yet measuredGabriel et al. (2024) Study 2 found personalized explanations are rated as more helpful than generic ones when well-aligned with users' attributes, but misaligned personalizations reduce effectiveness and can make users more resistant to accepting correction.
“Results indicate that personalized explanations are rated as more helpful than generic ones, particularly when well-aligned with users' attributes. However, misaligned personalizations reduce effectiveness.”
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 Gabriel et al. 2024
Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measuredGabriel et al. (2024) Study 1 found explanation-based interventions outperform label-only approaches, with GPT-4 explanations yielding the highest accuracy improvement (up to 47.6%) and the most significant reduction in misinformation sharing.
“Findings show that explanation-based interventions outperform label-only approaches, with GPT-4 explanations yielding the highest accuracy improvement (up to 47.6%) and the most significant reduction in misinformation sharing.”
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 Gabriel et al. 2024
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measuredChatGPT-4.0-generated personalized misinformation was difficult to detect, especially when aligned with participants' demographics.
“They found that ChatGPT-4.0-generated personalized misinformation was difficult to detect, especially when aligned with participants' demographics.”
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 Gabriel et al. 2024
LLMs can generate highly credible misinformation, particularly when personalized or tailored to user identities.
“First, LLMs can generate highly credible misinformation, particularly when personalized or tailored to user identities (Gabriel et al. 2024; Kim et al. 2024; Menz et al. 2024).”
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 Gabriel et al. 2024
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Learning Transferable Visual Models From Natural Language Supervision
Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. (2021) · arXiv · doi:10.48550/arxiv.2103.00020 · Cited by 15,293 · see citing works
The spread of true and false news online
Vosoughi, S., Roy, D., & Aral, S. (2018) · Science · doi:10.1126/science.aap9559 · Cited by 8,847 · see citing works
Flesch, R. (1948) · Journal of Applied Psychology · doi:10.1037/h0057532 · Cited by 5,391 · see citing works
Shifting attention to accuracy can reduce misinformation online
Pennycook, G., Epstein, Z., Mosleh, M., Arechar, A. A., Eckles, D., & Rand, D. G. (2021) · Nature · doi:10.1038/s41586-021-03344-2 · Cited by 1,155 · see citing works
Pennycook, G., Bear, A., Collins, E. T., & Rand, D. (2020) · Management Sciences · doi:10.1287/mnsc.2019.3478 · Cited by 689 · see citing works
Clayton, K., Blair, S., Busam, J. A., Forstner, S., Glance, J., Green, G., Kawata, A., Kovvuri, A., Martin, J., Morgan, E., et al. (2020) · Political Behavior · doi:10.1007/s11109-019-09533-0 · Cited by 658 · see citing works
Defending Against Neural Fake News
Zellers, R., Holtzman, A., Rashkin, H., Bisk, Y., Farhadi, A., Roesner, F., & Choi, Y. (2019) · arXiv · doi:10.48550/arxiv.1905.12616 · Cited by 569 · see citing works
Zhou, J., Zhang, Y., Luo, Q., Parker, A. G., & De Choudhury, M. (2023) · Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3581318 · Cited by 288 · see citing works
Effects of Credibility Indicators on Social Media News Sharing Intent
Yaqub, W., Kakhidze, O., Brockman, M. L., Memon, N., & Patil, S. (2020) · Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3313831.3376213 · Cited by 171 · see citing works
A Practical Guide to Doing Behavioral Research on Fake News and Misinformation
Pennycook, G., Binnendyk, J., Newton, C., & Rand, D. G. (2021) · Collabra Psychology · doi:10.1525/collabra.25293 · Cited by 164 · see citing works
Effects of fact-checking social media vaccine misinformation on attitudes toward vaccines
Zhang, J., Featherstone, J. D., Calabrese, C. R., & Wojcieszak, M. (2020) · Preventive Medicine · doi:10.1016/j.ypmed.2020.106408 · Cited by 149 · see citing works
We Can Detect Your Bias: Predicting the Political Ideology of News Articles
Baly, R., Martino, G. D. S., Glass, J. R., & Nakov, P. (2020) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.404 · Cited by 142 · see citing works
An Explanation of In-context Learning as Implicit Bayesian Inference
Xie, S. M., Raghunathan, A., Liang, P., & Ma, T. (2021) · arXiv · doi:10.48550/arxiv.2111.02080 · Cited by 138 · see citing works
Can LLM-Generated Misinformation Be Detected?Preprint
Chen, C., & Shu, K. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2309.13788 · Cited by 135 · see citing works
Bayesian or biased? Analytic thinking and political belief updating
Tappin, B. M., Pennycook, G., & Rand, D. G. (2020) · Cognition · doi:10.1016/j.cognition.2020.104375 · Cited by 127 · see citing works
Whose Opinions Do Language Models Reflect?
Santurkar, S., Durmus, E., Ladhak, F., Lee, C., Liang, P., & Hashimoto, T. (2023) · arXiv · doi:10.48550/arxiv.2303.17548 · Cited by 114 · see citing works
Language Models as Agent Models
Andreas, J. (2022) · Findings of the Association for Computational Linguistics: EMNLP 2022 · doi:10.18653/v1/2022.findings-emnlp.423 · Cited by 92 · see citing works
Self-citationDetecting Cross-Modal Inconsistency to Defend Against Neural Fake News
Tan, R., Plummer, B. A., & Saenko, K. (2020) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.163 · Cited by 62 · see citing works
Do Explanations Increase the Effectiveness of AI-Crowd Generated Fake News Warnings?
Epstein, Z., Foppiani, N., Hilgard, S., Sharma, S., Glassman, E. L., & Rand, D. G. (2022) · Proceedings of the International AAAI Conference on Web and Social Media · doi:10.1609/icwsm.v16i1.19283 · Cited by 61 · see citing works
Misinfo Reaction Frames: Reasoning about Readers’ Reactions to News Headlines
Gabriel, S., Hallinan, S., Sap, M., Nguyen, P., Roesner, F., Choi, E., & Choi, Y. (2021) · Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2022.acl-long.222 · Cited by 60 · see citing works
Self-citationExploring the Use of Personalized AI for Identifying Misinformation on Social Media
Jahanbakhsh, F., Katsis, Y., Wang, D., Popa, L., & Muller, M. (2023) · Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3581219 · Cited by 42 · see citing works
A Model of Online Misinformation
Acemoğlu, D., Ozdaglar, A., & Siderius, J. (2021) · National Bureau of Economic Research · doi:10.3386/w28884 · Cited by 39 · see citing works
Self-citationBayesian or biased? Analytic thinking and political belief updatingPreprint
Tappin, B. M., Pennycook, G., & Rand, D. G. (2019) · doi:10.31234/osf.io/a6euj · Cited by 33 · see citing works
Is Explanation the Cure? Misinformation Mitigation in the Short Term and Long Term
Hsu, Y.-L., Dai, S.-C., Xiong, A., & Ku, L. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.92 · Cited by 7 · see citing works
Danry, V., Pataranutaporn, P., Epstein, Z. G., Groh, M., & Maes, P. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2210.08960 · Cited by 5 · see citing works
Flagged first, then by citations
Research works printed but not linked: 2
- Radford, Alec, Jeff Wu, Rewon Child, David Luan, Dario Amodei and Ilya Sutskever. "Language Models are Unsupervised Multitask Learners." Preprint, 2019. https://paperswithcode.com/paper/language-models- are-unsupervised-multitask.
- Seo, Haeseung, Aiping Xiong, and Dongwon Lee. 2019. "Trust It or Not: Effects of Machine-Learning Warnings in Helping Individuals Mitigate Misinformation." In Proceedings of the 10th ACM Conference on Web Science, 265-74. WebSci '19. New York: Association for Computing Machinery.
Cited by
Sorted by citations
Park, S., & Nan, X. (2026) · AI & SOCIETY · doi:10.1007/s00146-025-02620-3 · Cited by 58 · see citing works
“Personalized corrections can improve discernment, yet misalignment with user identity or ideology may backfire (Gabriel et al. 2024), while model safeguards are neither durable nor uniformly effective (Menz et al. 2024; Węcel et al. 2023).”
InfluentialDesigning Conversational AI for Aging: A Systematic Review of Older Adults' Perceptions and Needs
Huang, Y.-H., Zhou, Q., & Piper, A. M. (2025) · International Conference on Human Factors in Computing Systems · doi:10.1145/3706598.3713578 · Cited by 54 · see citing works
Generative artificial intelligence and social media: insights for tobacco control
Kong, G., Ouellette, R. R., & Murthy, D. (2024) · Tobacco Control · doi:10.1136/tc-2024-058813 · Cited by 11 · see citing works
García‐Rudolph, A., David Sanchez-pinsach, B., Wright, M. A., Opisso, E., & Vidal, J. (2025) · Medical Teacher · doi:10.1080/0142159x.2024.2430365 · Cited by 4 · see citing works
The Influence of Artificial Intelligence on Communication
Talmaţchi, M.-L. (2025) · Scientific Bulletin · doi:10.2478/bsaft-2025-0014 · Cited by 3 · see citing works
Jamieson, J., Hara, T., & Akiyama, M. (2025) · CHI Extended Abstracts · doi:10.1145/3706599.3719948 · Cited by 3 · see citing works
The Role of AI Chatbots in Facilitating Online Harm: A Systematic Review
Amidu, G. (2025) · doi:10.21203/rs.3.rs-8427928/v1 · Cited by 1 · see citing works
Role of Generative AI in Fake News Creation and Detection
Barik, D., Guha, S. K., & Phani, S. (2026) · Lecture notes in networks and systems · doi:10.1007/978-981-96-9275-0_23
A risk-analysis framework for evaluating the impact of information disorders
Osman, M. (2026) · Journal of Risk Research · doi:10.1080/13669877.2026.2641506
AI Generative Tools: Redefining Learning for Malaysian University Students
Sinnappan, G. S. (2025) · International Conference on Emerging Innovation and Digital Technology (ICEIDT) · doi:10.1109/iceidt66693.2025.11473500
Richter, M. (2026) · Springer eBooks · doi:10.1007/978-3-032-16800-9_10
- Authors
- Gabriel, S., Lyu, L., Siderius, J., Ghassemi, M., Andreas, J., & Ozdaglar, A.
- Affiliations
- University of California, Los Angeles (US)
- Massachusetts Institute of Technology (US)
- Dartmouth College (US)
- MIT Computer Science and Artificial Intelligence Laboratory (US)
- MIT LIDS
- Venue
- An MIT Exploration of Generative AI
- Publisher
- PubPub
- Published
- 2024-03-27
- OpenAlex
- W4393231413
- Cited by
- 11
- References
- 26
- Citations by year
- 2026: 3 · 2025: 7 · 2024: 1
- FWCI
- 7.89
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by-nc
- Funders
- Dartmouth College
- Topics
- Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
- Physical Sciences › Computer Science › Information Systems › Spam and Phishing Detection
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Keywords
- generative AI, misinformation, disinformation, large language models, misinformation interventions, confirmation bias, personalized explanations, news reliability, alternative facts, social media, deficit model, critical thinking
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
