NEW YORK—As more government regulators and digital platforms start to require disclosure of AI-generated content, a new study from MediaScience found that labeling a video ad as AI-generated does not hurt how it performs.
The results showed no decline in any performance measure and increase in AI creation awareness across all four labeling conditions tested. For well-made ads, disclosure is not the threat the industry assumed it was, the researchers reported.
The study comes at a time when the regulatory pressure around AI in video advertising is ramping up.
New York's AI Transparency in Advertising law takes effect June 2026. The EU AI Act introduces binding disclosure obligations in August 2026.
These developments raised important questions for advertisers, who have been wondering if compliance and labelling would impact ad performance.
To address those issues, the study tested four labeling approaches across 900 U.S. respondents, reflecting frameworks under active consideration by U.S. and EU legislators: a text label in the first three seconds of the ad, a delayed text label from seconds four through six, a full-duration text label and a full-duration icon. Each was tested against a control with no labeling.
The data shows no adverse change in ad performance across any of the labeling conditions. Brand choice, ad memory, brand recognition, brand attitude, ad liking, and perceived production quality all showed no significant difference from the no-label control, the MediaScience survey found.
"There has been a lot of anxiety in the industry about what happens when you tell people an ad was made with AI," said Dr. Duane Varan, CEO of MediaScience. "The data gives us a clear answer: if the creative is good, disclosure does not hurt it. Advertisers do not need to be afraid of the label."
Displaying a disclaimer during the first three seconds increased viewers' awareness that the content was AI-generated by 28%. Running the label continuously throughout the ad increased awareness by 36%.
While 42% of respondents preferred the visual icon, it was the least effective at increasing AI awareness. On ad memory, text labels outperformed the control score of 36 across all conditions: 46 for the 3-second label, 40 for the delayed label, and 49 for full-duration. The icon scored 38, near the control.
The study also found that audiences feel the strongest need for AI labeling when it generates humans (60%), followed by animals (46%), product placement (45%), and voices (45%).
The study was conducted by MediaScience in collaboration with the Ehrenberg-Bass Institute at Adelaide University, a marketing science academic center, and MediaPET.ai, an AI video content platform developed by MediaScience.
