Synthetic-media detection vendor Resemble AI scanned brand ads in Meta's Ad Library and flagged 3,379 of 37,768 image and video files as AI-generated or manipulated. Only two of those flagged files mentioned AI anywhere in the ad copy. The results come from the vendor's own detector, but they land as AI disclosure rules tighten in New York and the EU.
What the study found
Resemble AI's 2026 Synthetic Advertising Report, published October 22 and covered by PPC Land on October 27, searched 509 companies across 26 industries in Meta's Ad Library during September 2026 and ran every image and video through its Resemble Detect model.
Overall, 8.9% of files were flagged. On brands' own pages the rate was 6.9%, and 222 of 389 brands ran at least one flagged file. So-called bait pages using brand names (clickbait, arbitrage and scam pages) were flagged at 71.3%.
- 70% of labelled fakes were photoreal; 46% showed a person
- Consumer packaged goods led brand categories at 13.4%
- Betting and crypto brand pages: 15.8% of files flagged
- 57 of 67 audio-flagged files were synthetic voices
- None of 971 files checked carried C2PA content credentials
The disclosure gap
Only 2 of the 3,379 flagged files, about 0.06%, mentioned AI in the ad copy. The report only checked on-screen labels for one advertiser, so visual labels elsewhere were not measured.
The regulatory backdrop is shifting. The report notes New York's synthetic performer disclosure rule has applied since June 9, 2026, and EU AI Act Article 50 transparency duties since August 2, 2026. Meta has labelled AI content as "Made with AI" since 2024 and signed the EU's code of practice on AI content transparency in July 2026.
Read it with care
Every verdict comes from one vendor's detector, with no published false-positive rate or outside validation, and PPC Land notes some totals in the report do not reconcile. Flagged testimonial-style clips are detector scores, not findings of wrongdoing. The report has no comment from Meta and no spend or reach data.
What it means for media buyers
AI avatars, synthetic voiceovers and AI "customer" testimonials are now common in Meta ads, and they are exactly the formats regulators and Meta's review systems are looking at. An undisclosed synthetic person presented as a real customer is the riskiest version, both for ad rejections and for account quality.
If you run AI-generated people or voices, disclose them where the law or Meta's policy requires it, keep your source files, and avoid presenting synthetic testimonials as real customer reviews. When an account does take a hit, having a supplier with a 24/7 operator, such as Power Ads, shortens the time to get a compliant replacement running.
What it means for you
- Audit live creatives for AI people, voices and testimonials, and label them where required
- Never present a synthetic person as a real customer in ad copy or video
- Keep originals and generation records in case a reviewer or regulator asks
- Treat the numbers as one vendor's estimate, not a Meta enforcement signal
