Hameleers, M., & van der Meer, T. (2026). Beyond textual disinformation: Comparing the effects of textual disinformation to AI-generated and video-based visual disinformation across different issues. New Media & Society. https://doi.org/10.1177/14614448251409208
Summary
This pre-registered experiment (N=982, U.S.) directly compares the effects of textual disinformation against two forms of visual disinformation—AI-generated still images and decontextualized real videos—across two issues (the disappearance of flight MH370 and the Russian invasion of Ukraine). The core argument pushes back against alarmist narratives about generative AI: visual disinformation does not universally outperform text in credibility, emotional arousal, or engagement. Instead, its persuasive power is context-bound, depending on issue salience, polarization, and—crucially—the availability of plausible authentic footage that can be recontextualized. Only for the more salient, polarized Ukraine war did decontextualized video prove more credible and engaging than text, while low-tech video manipulation often outperformed AI-generated imagery.
Key Contributions
- One of the first direct experimental comparisons of textual, AI-generated still-image, and decontextualized video disinformation within a single design.
- Demonstrates that visual disinformation effects are contingent on issue salience, polarization, and the availability of plausible authentic footage for recontextualization.
- Challenges alarmist framing of generative AI by showing that low-tech video decontextualization can be more persuasive than AI-generated imagery.
- Shows fact-checks are equally effective across modalities—fact-checkers need not build modality-specific correction tools, though they should explain the techniques behind visual deception.
- Clarifies moderating roles of conspiracy mentality, media distrust, and self-perceived media literacy across modalities.
Methods
- Pre-registered between-subjects online experiment (AsPredicted 6MX_9VL): 3 (modality: text vs. AI image vs. decontextualized video) × 2 (correction present/absent) + 3 controls, with topic (MH370 vs. Ukraine) varied within-subjects in randomized order.
- N=982 U.S. adults via Kantar opt-in panel with soft quotas (age, gender, education); collected December 5–11, 2023.
- Stimuli mimicked online news articles (MH370: false missile-strike claim; Ukraine: false Putin–Zelenskyy peace deal), each paired with decontextualized footage or AI-generated imagery.
- DVs: credibility (7 items), engagement intentions (6 items), discrete emotions (8 items), all 7-point scales. Moderators (media distrust, conspiracy mentality, self-perceived media literacy) measured pre-exposure.
- Analyses: ANOVAs with LSD post hoc, OLS regressions with two-way interactions, Wald coefficient comparisons, plus exploratory ideology and partisan-belief analyses.
Findings
- MH370: no significant modality differences in credibility, emotions, or engagement; all disinformation rated less credible than the authentic control.
- Ukraine: video disinformation was significantly more credible and engaging than text; differences from AI imagery were smaller or marginal.
- Emotional responses did not differ by modality for either issue.
- Conspiracy mentality amplified MH370 disinformation credibility across modalities (but not Ukraine), consistent with epistemic uncertainty driving susceptibility.
- Media distrust modestly increased credibility of textual (and somewhat video) MH370 disinformation.
- Self-perceived media literacy generally did not moderate effects, except reducing engagement with textual MH370 disinformation.
- Fact-checks significantly reduced credibility for both issues, with no interaction by modality.
- Exploratory: right-leaning participants were less responsive to some messages; stronger issue-specific partisan beliefs increased perceived credibility.
Connections
This paper’s empirical rebuttal to generative-AI alarmism complements experimental work finding limited persuasive advantages for AI-driven persuasion and disinformation, such as Hackenburg2025-dj and DeVerna2025-dl. Its focus on fact-check efficacy and resilience across contexts connects to comparative and correction-oriented studies including Cazzamatta2026-lo, and its treatment of decontextualized visual manipulation and AI-generated imagery relates to Di-Domenico2026-zq.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify · Apple Podcasts