AI disclosure labels show no downside in ad metrics, MediaScience study finds
MediaScience tested four AI disclosure formats in video ads. Results show stable recall and recognition, with text labels improving AI awareness.
AI labeling in video ads does not appear to reduce key brand and ad performance metrics, based on results from MediaScience’s “AI Labeling Impact Study” across 900 U.S. participants.
The study’s timing matters because disclosure requirements are expanding, including New York’s law focused on “synthetic performers,” and the EU AI Act’s labeling obligations for select AI-generated content.
Table of contents
Jump to each section:
- What the study tested and why it matters now
- What the numbers say across awareness, recall, and attitude
- Where consumers draw the line on when disclosure is needed
- What this means for marketers
What the study tested and why it matters now
MediaScience ran the research in conjunction with MediaPet (its AI video content platform) and the Ehrenberg-Bass Institute for Marketing Science at Adelaide University.
The experiment compared four disclosure methods against an unlabeled control group:
- Text label in the first three seconds
- Text label from seconds four through six
- Text label for the entire duration
- Icon displayed for the entire ad
These approaches were designed to mimic the kinds of disclosure frameworks being considered by EU and U.S. legislators, while giving advertisers a practical question to answer: does an AI label change how people respond to an ad?
ContentGrip
What the numbers say across awareness, recall, and attitude
On AI awareness, continuous text labeling produced the highest recognition that the ad used AI, with 49% of respondents identifying it as AI-generated. The unlabeled control group landed at 36%, while the icon method reached 38%. A label in the first three seconds reached 46%, compared with 40% when disclosure text appeared later (seconds four through six).
On unaided brand recall, the spread across groups was relatively tight (a seven-point difference), and some disclosure methods coincided with higher recall. The control group and icon group each recorded 54% unaided recall, while the seconds four through six group and continuous text group each reached 61%. The first-three-seconds label group recorded 60%.
Brand recognition results also varied within a narrow range (five points). Continuous text had the lowest recognition rate (87%), while the first-three-seconds label group reached the highest (92%). Control, seconds four through six, and icon groups recorded 88%, 91%, and 91% respectively.
Brand attitude was the area where the icon underperformed most clearly. Only 44% of participants in the icon group reported a positive brand attitude, compared with 51% for both the control group and the first-three-seconds label group. The seconds four through six and continuous text groups each recorded 49%.
Ad liking followed a similar pattern. The icon group reported 56% ad enjoyment, versus 63% in the control group. Text labels, whether partial or continuous, produced higher ad enjoyment: 70% for both short-duration text label groups, and 69% for continuous text.
Where consumers draw the line on when disclosure is needed
The study also measured when audiences believe AI labeling should be used, and the results point to a hierarchy of perceived sensitivity.
Consumers expressed the strongest preference for disclosure when AI is used to simulate humans (60%). That level dropped to 46% for animals. For product placement and voices, 45% of participants said disclosure is needed. Lower-sensitivity categories included animation (41%), background (35%), script writing (33%), music (28%), text/subtitles (24%), translation/dubbing (23%), and coloring and lighting (21%).
One practical mismatch emerged: icons were preferred by consumers over other disclosure styles by 13 percentage points, but icons were also the least effective method at increasing AI awareness (38%). Continuous text performed best on awareness (49%), while short, early text performed nearly as well (46%).
For marketers, that gap matters because compliance is not just about having a disclosure element, but about whether the disclosure is actually understood.
What this means for marketers
AI disclosure is becoming a design constraint inside creative, not just a legal line item. The study suggests the constraint is manageable, but disclosure choices can still shape comprehension and brand response.
- Treat disclosure format as a communication problem, not a checkbox
If the goal is genuine understanding that AI was used, text labels outperformed icons on awareness. That matters in jurisdictions where regulators may expect disclosures to be meaningful, not merely present. - If you must use an icon, plan for education or reinforcement
The icon approach had low awareness lift relative to the control group. If brands standardize around icons (because consumers say they prefer them), they may still need supporting cues that clarify what the icon signifies. - Front-loaded labeling can preserve outcomes while improving clarity
The first-three-seconds text label delivered strong awareness (46%) while staying within the tighter performance ranges seen across recall and recognition. That can be useful for teams trying to reduce risk while meeting disclosure expectations. - Segment labeling decisions by “synthetic human” risk, not by tool usage
Consumers’ strongest disclosure expectations clustered around simulated humans, with steep drop-offs for other AI use cases like lighting, translation, or subtitles. Marketers can use that gradient to prioritize where to add clearer disclosures and reviews. - Use measurement to choose the least-disruptive compliant pattern
The study’s most actionable insight is that disclosure did not show a negative impact across the tested performance measures. Teams can A/B disclosure timing and format while keeping an eye on brand attitude and ad liking, where the icon approach looked weaker.
Over time, disclosure will likely become part of brand trust mechanics, similar to how “sponsored” labels and privacy notices became standardized. The operational takeaway is that creative teams should design disclosures early, not retrofit them at the end of production.
The strategic takeaway is that disclosure can be positioned as a clarity feature rather than an apology, especially if the underlying creative is strong. The more disclosure becomes normalized, the more differentiation shifts back to the fundamentals: idea quality, brand fit, and execution.
