Ad platforms are rewriting your ads after sign-off

Your ad, but remixed

On 20 October, OpenAI emailed advertisers to say it would switch on “creative enhancements” for new and existing self-serve ChatGPT campaigns from 30 October. Search Engine Roundtable published the email. The feature uses the original ad and the landing page to tailor the ad to the user’s conversation or interests.

In practice, that means new or rewritten headlines and descriptions, copy translated into other languages, resized images and clicks sent to a different page on the advertiser’s site. Versions can be made in advance or on the fly. In OpenAI’s words, these ads “can run automatically without requiring you to review or approve each version.”

Advertisers can turn it off for the whole account or for individual campaigns in Ads Manager, and earlier opt-outs carry over. But according to PPC Land, a single switch covers all four kinds of change, so it is all or nothing.

It’s not just one platform

AI edits switched on by default aren’t new. In late October, Adweek reported on advertisers pulling ads on Facebook and Instagram after AI features they never asked for had changed them. Bookshop.org said an automatic account update added an AI illustration to its human-made ad without its knowledge. The agency Exverus said it had pulled live ads to remove sitelinks, translations and AI images or carousels.

Ryan Adams of Exverus told Adweek that many of the options sit in collapsed menus, and “if every one of those isn’t manually unchecked, the enhancements still apply.” So the direction of travel is clear, no cap: platforms want to optimise your client’s creative live, in real time. They are reachmaxxing. Your client is reputationmaxxing. Those aren’t always the same thing.

Who owns the AI’s mistakes is still a whole debate

The bigger question is who is responsible when an AI edit goes wrong. Adweek reported on 22 October that Meta’s revised advertising terms would make advertisers “solely responsible” for whatever its AI features create or change, including features switched on by default. Under the earlier wording, as Adweek described it, that risk applied only when advertisers chose to use the AI tools.

Those terms did not take effect. After complaints, mostly from publishers about a separate clause on crawling business websites, Meta delayed them, Reuters reported on 23 October. On 28 October, a Norwegian court suspended the terms there at the request of 180 media companies, and Meta then said its existing terms would stay in force while it consults industry groups, according to the Associated Press. On 30 October, the UK’s Competition and Markets Authority opened an investigation into whether the terms impose unfair conditions on publishers and advertisers. It has made no finding, and Meta said there were “no grounds” for one.

So the liability question is paused, not answered. And the edits themselves are a product setting, not a contract clause. They carry on regardless.

Why this lands on the account team’s desk

Agencies sell sign-off. The client approves a headline, a visual and a landing page, and the agency makes sure that is what runs. Default-on AI edits quietly break that promise. What runs might be a translation nobody checked, a headline nobody wrote or a landing page nobody chose.

That is a brand problem before it is a legal one. As Emarketer’s Asa Hiken told Adweek, it’s the brand that stands to lose, because people will assume it chose to use AI. In regulated categories, a rewritten claim or a translated price is the kind of thing a compliance team will not find cute.

And it is an awkward conversation. If a client asks who approved the version that ran, “the platform did it” is a red flag answer, especially if the setting was sitting in a menu the team never opened. Anyone who has been on the account side knows that question arrives at the worst possible moment, usually with a screenshot attached.

What account, media and creative teams can do now

None of this needs panic, and AI versions can be useful when they are planned. It just needs a bit of housekeeping before the Christmas and Black Friday rush makes it impossible.

• Audit every client account on every platform you buy. List which AI enhancements are on, at account, campaign and ad level, and screenshot the settings with a date.
• Agree a default with each client in writing: which enhancements are allowed, which are off and who can change that. Make it part of the media plan, not a side quest.
• Recheck after platform updates. Bookshop.org says an automatic account update is how its ad was changed, so a one-off audit is not enough.
• Check what actually ran. Where a platform shows the variations it served, review them as you would any creative, and keep a record.
• Brief creative with variation in mind. If a client wants AI versions, give the platform approved headlines, images and landing pages to work from, rather than letting it improvise.
• Ask your legal or commercial lead how your client contracts handle platform changes to approved work, so you have an answer before the question comes up.

What’s still unknown, and the vibe check

Plenty. It isn’t clear how much reporting platforms will give on which AI versions of an ad ran and where. The paused terms may return in a different form, and the UK investigation has only just started. Platforms will also keep adding features, and some will arrive switched on.

So the vibe for agencies is curious but in control. AI versions might well improve results, and some clients will want them. The point is that it should be a choice made with the client, not a default nobody noticed. The agencies that come out of this well will be the ones that can show a client exactly what was approved and what ran, with receipts.