Labeling AI-generated media online
The DOI matches its Crossref record; 18 of the 18 references it cites were found in a registry or an index.
Abstract
Recent advancements in generative AI have raised widespread concern about the use of this technology to spread audio and visual misinformation. In response, there has been a major push among policymakers and technology companies to label AI-generated media appearing online. It remains unclear, however, what types of labels are most effective for this purpose. Here, we evaluate two (potentially complementary) strategies for labeling AI-generated content online: (i) a process-based approach, aimed at clarifying how content was made and (ii) a harm-based approach, aimed at highlighting content's potential to mislead. Using two preregistered survey experiments focused on misleading, AI-generated images (total n = 7,579 Americans), we assess the consequences of these different labeling strategies for viewers' beliefs and behavioral intentions. Overall, we find that all of the labels we tested significantly decreased participants' belief in the presented claims. However, in both studies, labels that simply informed participants that content was generated using AI tended to have little impact on respondents' stated likelihood of engaging with their assigned post. Together, these results shed light on the relative advantages and disadvantages of different approaches to labeling AI-generated media online.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
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
“Depending on the user and the content, the same label can make a video seem unusual, prompt a moment of curiosity, or introduce a reason to pause and verify (Van der Bend et al., 2023; Wittenberg et al., 2025).”
From AI-Generated Content Disclosure and Prolonged Short-Video Engagement: A Heuristic-Systematic Risk-Trust Model Among Late-Adolescent and Emerging-Adult TikTok Users (2026) · cites this paper as Wittenberg et al., 2025
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Cognitive Reflection and Decision Making
Frederick, S. (2005) · Journal of Economic Perspectives · doi:10.1257/089533005775196732 · Cited by 5,256
Prior exposure increases perceived accuracy of fake news
Pennycook, G., Cannon, T. D., & Rand, D. G. (2018) · Journal of Experimental Psychology General · doi:10.1037/xge0000465 · Cited by 1,268
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
Self-citationArt and the science of generative AI
Epstein, Z., Hertzmann, A., the Investigators of Human Creativity, Akten, M., Farid, H., Fjeld, J., Frank, M. R., Groh, M., Herman, L., Leach, N., et al. (2023) · Science · doi:10.1126/science.adh4451 · Cited by 788 · cited 2×
“Yet much of the AI-generated content in circulation today is meant for enter-tainment and creative expression, with little intent or capacity to deceive wide swaths of the public ( 70).”
Self-citationClayton, 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
All the News That’s Fit to Fabricate: AI-Generated Text as a Tool of Media Misinformation
Kreps, S., McCain, R. M., & Brundage, M. (2022) · Journal of Experimental Political Science · doi:10.1017/xps.2020.37 · Cited by 360 · cited 1×
“…with the concept of AI (3, 24), many people report limited exposure to and direct experience with generative AI technologies (25, 26) and often struggle to discern whether images, audio, and video have been created using AI (e.g. (27 – 30); for studies of AI-generated text, see also (31, 32)).”
AI model GPT-3 (dis)informs us better than humans
Spitale, G., Biller-Andorno, N., & Germani, F. (2023) · Science Advances · doi:10.1126/sciadv.adh1850 · Cited by 241 · cited 1×
“…with the concept of AI (3, 24), many people report limited exposure to and direct experience with generative AI technologies (25, 26) and often struggle to discern whether images, audio, and video have been created using AI (e.g. (27 – 30); for studies of AI-generated text, see also (31, 32)).”
Timing matters when correcting fake news
Brashier, N. M., Pennycook, G., Berinsky, A. J., & Rand, D. G. (2021) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2020043118 · Cited by 216 · cited 1×
“…has also shown previous success in related settings; labeling content that fact-checkers have flagged as false generally reduces viewers’ belief in (33 – 38) and likelihood of sharing (36, 39 – 41) this content, even in cases where people are pre-disposed to discount labels as biased ((42); for a…”
Self-citationSundar, S. S., Molina, M. D., & Cho, E. (2021) · Journal of Computer-Mediated Communication · doi:10.1093/jcmc/zmab010 · Cited by 212 · cited 1×
“But these established strategies may or may not work for AI-generated content (44, 45) — particularly audio and visual media (46, 47), which often seem especially credible and realistic (48, 49) and tend to attract outsized engagement (28).”
Fooled twice – people cannot detect deepfakes but think they can
Köbis, N., Doležalová, B., & Soraperra, I. (2021) · doi:10.2139/ssrn.3832978 · Cited by 194
Research works printed but not linked: 30
- Helmus, 2022, Artificial intelligence, deepfakes, and disinformation: a primer
- Marcinek, 2024, Risk-based AI regulation: a primer on the artificial intelligence act of the European Union
- Straub, 2022, Americans’ perspectives on online Media warning labels, Behavioral Sciences, 12, 59, 10.3390/bs12030059 · DOI
- DiResta, 2024, How spammers and scammers leverage AI-generated images on Facebook for audience growth, Harvard Kennedy School Misinformation Rev, 5, 10.37016/mr-2020-151 · DOI
- Koch, 2023, Effects of fact-checking warning labels and social endorsement cues on climate change fake news credibility and engagement on social media, J Appl Soc Psychol., 53, 495, 10.1111/jasp.12959 · DOI
- Pennycook, 2020, The implied truth effect: attaching warnings to a subset of fake news headlines increases perceived accuracy of headlines without warnings, Manage Sci., 66, 4944, 10.1287/mnsc.2019.3478 · DOI · Paper page
- Porter, 2022, Political misinformation and factual corrections on the Facebook news feed: experimental evidence, J Polit., 84, 1812, 10.1086/719271 · DOI
- Shen, 2021, Research note: this photograph has been altered: testing the effectiveness of image forensic labeling on news image credibility, Harvard Kennedy School Misinformation Rev, 2, 10.37016/mr-2020-72 · DOI
- Mena, 2020, Cleaning up social media: the effect of warning labels on likelihood of sharing false news on Facebook, Policy Internet., 12, 165, 10.1002/poi3.214 · DOI · Paper page
- Martel, 2024, Fact-checker warning labels are effective even for those who distrust fact-checkers, Nat Hum Behav., 8, 1957, 10.1038/s41562-024-01973-x · DOI
- Lewis, 2023, Deepfake detection with and without content warnings, R Soc Open Sci., 10, 231214, 10.1098/rsos.231214 · DOI
- Ternovski, 2022, The negative consequences of informing voters about deepfakes: evidence from two survey experiments, Journal of Online Trust and Safety, 1, 10.54501/jots.v1i2.28 · DOI · Paper page
- Oeldorf-Hirsch, 2020, The ineffectiveness of fact-checking labels on news memes and articles, Mass Commun Soc., 23, 682, 10.1080/15205436.2020.1733613 · DOI
- Morrow, 2022, The emerging science of content labeling: contextualizing social media content moderation, J Assoc Inf Sci Technol., 73, 1365, 10.1002/asi.24637 · DOI
- Li, 2024, Impact of artificial intelligence-generated content labels on perceived accuracy, message credibility, and sharing intentions for misinformation: web-based, randomized, controlled experiment, JMIR Form Res., 8, e60024, 10.2196/60024 · DOI · Paper page
- Peng, 2023, An agenda for studying credibility perceptions of visual misinformation, Polit Commun., 40, 225, 10.1080/10584609.2023.2175398 · DOI
- Newman, 2024, Misinformed by images: how images influence perceptions of truth and what can be done about it, Curr Opin Psychol., 56, 101778, 10.1016/j.copsyc.2023.101778 · DOI
- Mosleh, 2020, Self-reported willingness to share political news articles in online surveys correlates with actual sharing on Twitter, PLoS One, 15, e0228882, 10.1371/journal.pone.0228882 · DOI · Paper page
- Dobber, 2025, The effect of traffic light veracity labels on perceptions of political advertising source and message credibility on social media, J Inform Technol Politics, 22, 82, 10.1080/19331681.2023.2224316 · DOI
- Guo, 2023, Seeing is not believing: a nuanced view of misinformation warning efficacy on video-sharing social media platforms, Proc ACM Hum Comput Interact., 7, 1
- Burke, 2005, High-cost banner blindness: ads increase perceived workload, hinder visual search, and are forgotten, ACM Trans Comput Hum Interact., 12, 423, 10.1145/1121112.1121116 · DOI · Paper page
- Grady, 2021, Nevertheless, partisanship persisted: fake news warnings help briefly, but bias returns with time, Cogn Res Princ Implic., 6, 52, 10.1186/s41235-021-00315-z · DOI · Paper page
- Farid, 2022, Creating, using, misusing, and detecting deep fakes, J Online Trust Safety, 1, 10.54501/jots.v1i4.56 · DOI · Paper page
- Freeze, 2021, Fake claims of fake news: political misinformation, warnings, and the tainted truth effect, Polit Behav., 43, 1433, 10.1007/s11109-020-09597-3 · DOI · Paper page
- Schiff, 2025, The Liar's dividend: can politicians claim misinformation to evade accountability?, Am Polit Sci Rev., 119, 71, 10.1017/S0003055423001454 · DOI
- Watson, 1988, Development and validation of brief measures of positive and negative affect: the PANAS scales, J Pers Soc Psychol., 54, 1063, 10.1037/0022-3514.54.6.1063 · DOI · Paper page
- Cacioppo, 1982, The need for cognition, J Pers Soc Psychol., 42, 116, 10.1037/0022-3514.42.1.116 · DOI · Paper page
- Thomson, 2016, Investigating an alternate form of the cognitive reflection test, Judgm Decis Mak., 11, 99, 10.1017/S1930297500007622 · DOI · Paper page
- Schwartz, 2021, A repository of Schwartz value scales with instructions and an Introduction, Online Read Psychol Culture, 2, 9
- Wittenberg, 2025, 10.17605/OSF.IO/F8GQP · doi:10.17605/osf.io/f8gqp
Cited by
Altay, S., & Gilardi, F. (2024) · PNAS Nexus · doi:10.1093/pnasnexus/pgae403 · Cited by 119
Höltervennhoff, S., Ricker, J., Raphael, M. M., Schwedes, C., Weil, R., Fischer, A., Holz, T., Schönherr, L., & Fahl, S. (2026) · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3791006 · Cited by 3
Xiao, Y., Du, J., Ding, Y., Zhang, M., Jiang, Y., Yang, Y., & Liu, J. (2026) · Behavioral Sciences · doi:10.3390/bs16071179 · Cited by 2
Metadata
- Authors
- Wittenberg, C., Epstein, Z., Péloquin-Skulski, G., Berinsky, A. J., & Rand, D. G.
- Affiliations
- Sloan School of Management, Massachusetts Institute of Technology , Cambridge, MA 02142 ,
- Stanford Institute for Human-Centered AI, Stanford University , Palo Alto, CA 94305 ,
- Department of Political Science, Massachusetts Institute of Technology , Cambridge, MA 02139 ,
- Venue
- PNAS Nexus
- Publisher
- Oxford University Press (OUP)
- Published
- 2025-05-28
- Volume
- 4
- Issue
- 6
- ISSN
- 2752-6542
- Cited by
- 45
- References
- 18
- Language
- en
- Funders
- MIT Generative AI Impact Awards
- MIT Social and Ethical Responsibilities of Computing
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Labeling AI-generated media online. https://app.cochat.ai/papers/10.1093/pnasnexus/pgaf170
- The paper
- Wittenberg, C., Epstein, Z., Péloquin-Skulski, G., Berinsky, A. J., & Rand, D. G. (2025). Labeling AI-generated media online. PNAS Nexus. https://doi.org/10.1093/pnasnexus/pgaf170
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.