From ADOTATOpenAI and Meta Call It Enhancement. You're Liable for Ads You Never Read.
OpenAI and Meta Call It Enhancement. The Ad You Never Saw Is Still Your Liability.
On Oct. 20, OpenAI told advertisers its new creative enhancements can rewrite headlines and descriptions, translate copy, resize images and send the click to a different page on your site, all "without requiring you to review or approve each version." It is on by default. Meta switched on product browsing for running ads on Oct. 28, and its revised Commercial Terms took effect Oct. 30. Keep running ads and you have agreed.
The paper trail is the story. OpenAI's July 31 Ad Tools Terms said you were "responsible for reviewing Generated Creatives before use." The Aug. 24 version made you responsible for creatives "applied to your Campaign", extended your indemnity to them, and dropped the review sentence. The Oct. 15 Advertising Terms named features enabled by default. Then the default went on. That is not timing. That is a sequence. Meta's new AI Features section, as Adweek read it, makes advertisers solely responsible for what its AI creates or changes, including features Meta turns on itself, while conceding the AI "may make mistakes."
Why it matters: the platforms cap their own liability, take the indemnity, and grade the results. OpenAI's own help page says its standard numbers can't isolate what an enhancement did. Meta's 3% conversion lift comes from a 16-day August test with no median, no spread and no CPA. Agencies are already pulling ads to strip out edits, and REI pulled an Instagram ad in June after an AI-altered bike grew a second set of handlebars.
The fix is boring, which is why nobody does it. Turn it off at the account and the campaign level, screenshot it, and archive your approved copy today. A platform that can change your ad, cap its liability and grade its own work can never be wrong, and you can never prove it was.
Expanded from ADOTAT's Oct. 29 investigation, which compares OpenAI's Ad Tools Terms (July 31 U.S. and Aug. 24 UK versions), its Oct. 15 Advertising Terms, its help page and Oct. 20 advertiser email, Meta's product browsing notice as reproduced by PPC Land, and Adweek's reporting on Meta's revised Commercial Terms. Meta's full AI Features text was not available to ADOTAT, the 3% figure is Meta's own and unverified, and agency accounts are as given to Adweek or posted publicly. ADOTAT discloses that it uses AI tools, including Anthropic's Claude, and that Anthropic competes with OpenAI; the author does paid work for a law firm that defends companies in marketing-compliance cases.