This Video Was Made by AI—Does That Make You Click More or Less?

TikTok, Instagram reels, YouTube shorts—we’ve heard (and probably watched) them all. Short-form videos are the cultural bloodstream of the internet, and the influx of AI-generated content only raises the stakes. On one hand, AI dramatically lowers barriers to creation; on the other, it raises concerns about authenticity.

Research into this is especially timely. As governments push for transparency and platforms roll out AI-labeling policies, understanding how disclosure shapes behavior steers the future of online trust, creativity, and regulation. Policymakers have already begun to act. The European Union, for instance, now requires major platforms to label AI-generated images and videos.¹ These efforts rest on a common assumption that transparency helps users make better and more informed decisions, but does it sacrifice content engagement? 

Past studies across domains demonstrate our fractured relationship with AI usage. People trust medical diagnoses less when told they’re from AI.² Job applicants hesitate when recruiting systems reveal AI involvement.³ Even poetry written by AI tends to be judged more harshly than its human-authored counterpart.⁴

However, there is a key difference between these examples and short-form content. Such videos are fast, emotional, and guided by heuristics—gut feelings rather than careful reasoning. Could AI disclosure in such an environment play by different rules? 

The Heuristic-Systematic Model

Chen and colleagues investigate the reception of short-form content through the Heuristic–Systematic Model, which distinguishes between quick, effortless judgments and slower, more deliberate analysis. In short-form media, both processes fire at once, raising the possibility that disclosure triggers competing reactions. 

Examining User Engagement
On the Chinese survey platform Credamo, 479 participants were randomly assigned to evaluate a screenshot of a TikTok video introducing a product to the viewer. For half the participants, the screenshot contained a headline at the top explicitly stating that the content was AI-generated, while the other half received no information. 

Participants then rated perceived content quality, their intention to engage with the video, and their perceptions of AI’s capabilities measured in terms of reliability, flexibility, and integrability. The data was analyzed using partial least squares structural equation modeling (PLS-SEM), allowing them to examine both the direct and indirect pathways linking AI disclosure to user engagement.

The results of the modeling highlight a paradoxical relationship between disclosure, engagement, and perceived content quality. While disclosure increased user engagement, it also lowered perceived content quality. When viewed through the lens of the Heuristic-Systematic model, this suggests that the novelty of AI-generated content creates quick impressions that nudge people toward engagement without much deliberation. On the flip side, longer analysis leads users to think more systematically and critically about the content they are watching. This deeper processing may explain the drop in perceived quality—an effect strong enough to counteract the initial excitement. 

Crucially, the researchers found that perceptions of AI’s capabilities changed everything. When users rated AI as reliable, flexible, and integrated, disclosure had less of an effect on perceived content quality. In fact, at high levels of perceived AI capability, disclosure no longer decreased content quality perceptions at all—it actually increased it. If people think AI is competent, disclosure becomes an asset; if they doubt it, disclosure can backfire. 

The implication is subtle but important: disclosure doesn’t operate in a vacuum. Its impact depends on the assumptions users hold about AI itself. This reframes the challenge facing online platforms. Instead of minimizing AI’s presence, there may be value in making its capabilities more legible to users. Thoughtful interface cues or experiential design choices could help build AI literacy and, in turn, reduce the risk that disclosure undermines perceived quality.

One open question emerging from this study concerns the language of disclosure itself. Do users respond differently to labels like “AI-assisted” versus “AI-generated”? Distinguishing how disclosure wording shapes engagement and perceptions of quality could reveal whether language alone meaningfully alters how audiences interpret AI involvement. On top of that, the study paves new ground for future research to explore AI disclosure across different content types. Would these effects hold in political content? Entertainment? Deepfake scenarios? There is plenty of room for further analysis to come to a holistic understanding of transparency and its rippling effects.

The debate over AI transparency often assumes that disclosure is either good or bad. This study shows it’s both—and neither. The real story lies in the tug-of-war between intuition and analysis. When users see “AI-generated,” their curiosity pulls them in even as their skepticism pulls them back. What determines which way someone leans depends on what they believe AI is capable of. 

Chen and colleagues show us that AI disclosure isn’t inherently risky or beneficial. Its impact hinges on users’ confidence in the technology behind the label. In a world when machine-generated content is becoming indistinguishable from human creativity, the future of online trust won’t be decided by transparency alone, but by how people feel about the intelligence doing the creating.

1. European Parliament. (2023, June 1). EU AI Act: First regulation on artificial intelligence. https://www.europarl.europa.eu/topics/en/article/20230601STO93804/eu-ai-act-first-regulation-on-artificial-intelligence
2. Yokoi, R., Eguchi, Y., Fujita, T., & Nakayachi, K. (2021). Artificial intelligence is trusted less than a doctor in medical treatment decisions: Influence of perceived care and value similarity. International Journal of Human–Computer Interaction, 37(10), 981–990. https://doi.org/10.1080/10447318.2020.1861763
3. Keppeler, F. (2024). No thanks, dear AI! Understanding the effects of disclosure and deployment of artificial intelligence in public sector recruitment. Journal of Public Administration Research and Theory, 34(1), 39–52. https://doi.org/10.1093/jopart/muad009
4. Köbis, N., & Mossink, L. D. (2021). Artificial intelligence versus Maya Angelou: Experimental evidence that people cannot differentiate AI-generated from human-written poetry. Computers in Human Behavior, 114, Article 106553. https://doi.org/10.1016/j.chb.2020.106553