If you are running ads on ChatGPT, OpenAI sent you an email last week with a short fuse, and the fuse just burned down. Creative enhancements are now switched on by default for self-serve campaigns, new and existing. With them on, the system can write new headlines and descriptions, translate your ad text, resize your images, and send your clicks to a different page on your site than the one you picked. It can run those versions without you approving each one.
The rollout itself was messy: the email said Friday, October 30, while OpenAI's own help page said Thursday, October 29, with no explanation for the difference. Either way, as of this week the safe assumption is that it is on in your account unless you turned it off.
The Part That Matters
We use AI every day. We sell AI chat and AI voice, we test AI-written ad variants, and some of them beat the human versions. So this is not an anti-AI post. AI writing ad copy is fine.
What is not fine is the combination this ships with: no preview of every variation, and reporting that stops at the ad level. You may never see the exact headline a customer saw, and the reports will not tell you which version ran. So when lead quality shifts, and lead quality is the whole game, you cannot tell whether the rewrite caused it. You just see the blended number move and get to guess why. That is not a creative problem, it is a measurement problem, and measurement problems are the expensive kind.
There is also the claims problem. If the AI rewrites "licensed and insured, financing available" into something stronger, or translates your guarantee into a language nobody at your company reads, that claim is still yours. The ad runs under your name. No category safeguards have been described for regulated industries, and the published terms do not spell out who is liable for an AI rewrite. Until they do, assume you are.
What to Do Now
The toggle lives in the Settings section of Ads Manager, labeled Creative enhancement, on by default. It works at any time, not just before the deadline. You can turn it off at the account level or per campaign, and campaign settings override the account default. One catch: it is all four enhancement types or none. You cannot keep translation and drop the rewrites.
Our take: turn it off, then test it on purpose. Pick one campaign where you can watch lead quality closely, let the enhancements run, and compare booked work, not clicks, against a campaign you control word for word. If the AI versions win on the numbers that hit your bank account, great, expand it. That is testing. Leaving it on everywhere because a default flipped on a Thursday is not testing, it is drifting.
Defaults are not neutral. When a platform turns something on for everyone at once, it is because the feature serves the platform at scale. Sometimes it serves you too. The only way to know is to measure it like you measure everything else, all the way down to revenue.
That measurement is the part most accounts are missing. If your attribution stops at the click, you cannot tell an AI rewrite that books jobs from one that books tire-kickers. AdStack™ runs ChatGPT campaigns alongside Google and Meta with the tracking wired through to booked revenue, so a test like this one has a scoreboard. If you want to know what your enhanced ads have actually been saying, book a call.
Article imagery is illustrative. Product names, logos, and brands that may appear in images or text are the property of their respective owners and are used for identification and commentary only; their appearance does not imply any affiliation with, or endorsement by, those owners.
