Meta’s ad archive looks far more synthetic than the copy suggests. Resemble AI scanned 37,768 image and video files from Facebook and Instagram ads in September 2026 and flagged 3,379 as AI-generated or manipulated — about 8.9%, or one in eleven files.
A massive gap between creative and disclosure
The 2026 Synthetic Advertising Report found that of the 3,379 flagged files, only two mentioned AI anywhere in the ad copy — a 0.06% written-disclosure rate. The only brand-page example named was Turkish Airlines, whose International Cat Day video for its Touristanbul programme carried a footnote saying the visuals “were created using artificial intelligence technology.”
The audit drew on Meta’s public Ad Library and covered 509 companies across 26 industries. On brands’ own pages, 1,971 of 28,463 files were flagged, or 6.9%. Bait pages were far worse: 71.3% of 649 files, ten times the brand-page rate.
- 222 of 389 brands (57%) ran at least one AI file from their own page.
- 13 brands had more than 40% of sampled files flagged.
- Four large advertisers sat above half: Temu, Wayfair, Motel 6 and Lufthansa.
Why this matters now
New York’s synthetic performer disclosure law has applied since June 9, 2026, and Article 50 of the EU AI Act since August 2, 2026. Both put most of the duty on whoever makes or deploys the ad, not the platform carrying it. Yet none of the 971 files checked carried C2PA content credentials, and Resemble’s copy search could not see visual labels burned into video.
Most flagged creative was designed to pass as real. Of the 3,216 labelled files, 70% were photoreal and 46% showed a person. Resemble’s report puts the contrast bluntly: “Brands fake polish. Bait pages fake people.”
Read the fine print
This is a vendor study with a commercial interest: Resemble sells synthetic-media detection, and the report’s final page advertises its detection models. It also has no impression, reach or spend data, so a flagged file that reached a thousand people counts the same as one that reached ten million.
Audio is a blind spot in the main collection. The report says every rate is a floor: in the betting and crypto set, 62 of 142 fakes were caught on audio alone, and 55 synthetic voices were traced to ElevenLabs.
What marketers should do
Treat this as a creative inventory and trust problem. Log where generative AI enters your workflow, and audit photoreal people, selfie-style testimonials and synthetic voices. The IAB’s AI Transparency and Disclosure Framework, updated August 18, 2026, requires labels for photorealistic AI influencers and synthetic voices of real people, but exempts generic synthetic voiceovers and background music. Use that line to decide what needs a visible label.
Disclose clearly in the creative itself where the label changes what a consumer believes — not only in a caption footnote after hashtags. Also monitor bait pages that use your brand name as a lure. They made up 2% of all files but 14% of all fakes.
Source: PPC Land
