OpenAI is about to start rewriting the ads you run in ChatGPT. New headlines, new descriptions, translations into other languages, resized images and, the part that should make you sit up, a different landing page on your site if the system thinks it fits the conversation better. The versions can run without you seeing them first. The setting is on unless you turn it off.
This is a primary-source review. I haven’t run creative enhancements on a live account, and I haven’t found published results from any advertiser who has. What follows comes from OpenAI’s own help center, its new Ad Tools Terms, the advertiser email as published by two trade outlets, and OpenAI’s reporting documentation. That’s enough to make a decision this week, which is the point, because the deadline is next week.
What it is
Creative enhancements is an automated creative layer inside ChatGPT Ads Manager. It takes three inputs: “the context of the conversation or user interests,” your existing ad, and “content from the ad’s landing page.” From those it can change four things:
- Headlines and descriptions. OpenAI “may create new or updated text to connect your offering to the conversation or user interests.”
- Language. It “may translate your ad text into another language.”
- Images. They “may be resized to fit the ad format.”
- Destination URLs. The URL “may be optimized to direct people to a relevant page on your website.”
Variations “may be prepared in advance or created in real time based on the conversation.” And the sentence that matters most: “When enabled, these ads can run automatically without requiring you to review or approve each version.” The FAQ goes further: the experience “does not provide a preview of every variation.”
OpenAI’s own example is harmless enough. A grocery ad that says “Fresh groceries delivered” becomes “Groceries for quick weeknight dinners” in a conversation about weeknight meals. If every rewrite looked like that, nobody would care. The question is what it does with your mortgage rate, your supplement claim or your Black Friday price.
What changed this week
Two things, a day apart.
On October 20 OpenAI emailed advertisers. Search Engine Roundtable published it that afternoon, and Search Engine Land followed on October 21. The email says: “Starting Friday, October 30, 2026, we’ll enable creative enhancements for your new and existing self-serve campaigns.” The help page, as I read it on October 22, says something slightly different: “Starting Thursday, October 29, 2026, creative enhancements will be enabled automatically for new and existing campaigns.” One day earlier, and no “self-serve” qualifier. PPC Land caught the same mismatch on October 21. OpenAI hasn’t explained either difference. Plan for the earlier date and assume it covers every campaign until someone at OpenAI tells you otherwise in writing.
On October 21 OpenAI also published its Ad Tools Terms. Section 3 covers “AI Creative Tools,” and it’s the real document behind the toggle. The settings card in Ads Manager, per the screenshot Search Engine Roundtable published and PPC Land transcribed, says the feature is “Subject to the Ad Tool Terms.”
Context makes the timing sharper. The same day, OpenAI said ChatGPT Ads now runs in 89 countries and that more than 100,000 advertisers have used it. Ads show only to Free and Go users. Translation by default in 89 markets is a very different thing from translation in one.
What works
Credit where it’s due. A few things here are better than the usual platform default flip.
The control model is simple and sticky. There’s an account default and a campaign-level override, and the help page says “existing account- and campaign-level opt-outs are preserved” and “campaign-specific settings take priority over your account default.” If you turned this off already, you stay off. You can switch it back any time.
The notice was real. Ten days by email, nine by the help page. That’s short, but it’s more than some platforms give, and the email explained what changes in plain language.
The help page is honest about measurement. OpenAI writes that Ads Manager metrics “describe overall performance” and that “changes over time do not, on their own, establish the effect of creative enhancements.” That’s a platform telling you, in its own docs, not to trust a before-and-after chart. Take it at its word.
Translation could genuinely help small advertisers. A business with one English landing page and a product that sells abroad now has a way into conversations in other languages without paying for localization. OpenAI hasn’t published which languages are covered. If your offer is simple and your claims are low-risk, that’s real reach.
The rest of the toolset is still conservative. The Create Ads guide says the “suggested ad drafts” feature “does not generate new copy or imagery with AI.” OpenAI is drawing a line between drafts you approve and enhancements that run on their own, and saying which is which.
What breaks
One switch, four very different powers. Rewording a headline is creative optimization. Picking a different page on your site is a media decision. Translating your copy is a localization and compliance decision. Neither the help page nor the email describes separate controls, and PPC Land’s reading of the settings card is a single toggle covering all four. You can’t keep resizing and turn off URL changes.
No preview, no version reporting. The Measure Results page lists impressions, clicks, spend, CTR, average CPC and CPM, and conversions, at campaign, ad group and ad level, with Device, Country and Platform breakdowns. Nothing in it shows which headline, language or landing page a given impression used. You’ll be optimizing an ad whose words you can’t read.
You own what you never saw. This is the part your legal team needs to read. Under section 3.2 of the Ad Tools Terms, you’re responsible for “all Generated Creatives that you approve or use, or that are applied to your Campaign,” including “the accuracy, completeness, currency, and substantiation” of claims about pricing, availability, testimonials, performance, risks and expected results, plus any AI disclosure the law requires. Your indemnification obligations apply to them. Section 3.3 says OpenAI doesn’t guarantee generated creatives will be accurate, compliant with law or “consistent with your intended meaning,” and isn’t responsible for claims arising from them. Read together: OpenAI writes it, you answer for it.
You can’t reuse the winners. Section 3.1 says any applied generated creative is “deemed an Ad,” and section 3.4 says generated creatives “may not be used outside the Advertising Services unless expressly permitted by OpenAI.” So if the system finds a headline that converts, you can’t see it in reporting, and you may not be allowed to lift it into your search ads anyway.
Your whole site becomes source material and destination. The inputs include landing page content, and the URL can move to “a relevant page on your website.” Nothing in the help page describes a page exclusion list. Every stale promo page, expired price, old pricing table and half-finished microsite is now potential ad copy or a potential landing page. The help page’s advice is to keep “prices, promotions, product details, and availability” current. That’s a site governance job most marketing teams don’t own.
Tracking may not follow the click. OpenAI supports dynamic URL parameters with macros like {campaign_id} and {ad_id}, set at ad URL, ad, ad group or campaign level, with the ad URL level winning. What the docs don’t say is what happens to those parameters when OpenAI swaps the destination. If your tagging lives only on the ad URL, assume it could be lost when the page changes, and check.
Regulated categories get no special handling. OpenAI’s policy troubleshooting page lists categories allowed in the US only, such as legal services, insurance, deposit accounts and dietary supplements, and says ads go through creative and landing page review. Neither source says whether each generated variation goes through that review, or whether a translated version can serve in a market where your category isn’t allowed. Practitioners noticed fast. Melissa Mackey, replying on X on October 20, said she’d already paused a ChatGPT ads test over lead quality and was “glad we did.”
Who it’s for
Leave it on if you’re a small advertiser with one clean site, a simple evergreen offer, no regulated claims, and no in-house capacity to write ten headline variants. OpenAI’s own guidance says “a high volume and diverse set of ads” helps the system find more places your ad fits, and enhancements are a shortcut to that for people who’d never write the variants themselves.
Turn it off if you’re in finance, health, legal, insurance or anything with claims that need substantiation; if you run frequent promotions or price changes; if brand voice is a contractual matter with a licensor or retailer; if you sell in several countries and care what your ads say in each language; or if you’re an agency running client accounts, where the client carries the liability and you made the call.
Pricing and limits
Neither the help page nor the email mentions a separate fee. You pay through the same clicks and impressions, so the cost is indirect: if rewritten ads win more auctions in conversations you’d never have chosen, you pay for that traffic. The documented limits are the ones above: no per-variation preview, no per-variation reporting, no documented per-type control, no documented URL exclusions, terms that put liability on the advertiser, and a start date that differs by a day depending on which OpenAI source you read.
What to do this week
- Decide by Wednesday, October 28. Use the help page’s October 29 date, not the email’s October 30. Set the account default on purpose, either on or off, and write down who decided and why.
- If you test, test properly. Turn enhancements on for one campaign with an evergreen offer, a clean landing page and one target language. Run a matched campaign with them off. Compare the two over the same dates, because OpenAI itself says a before-and-after chart proves nothing.
- Audit your site as if it were ad copy. Find expired promotions, outdated prices and pages you’d never want as a landing page (careers, investor relations, support, legal). Fix, redirect or remove them before October 29. Check that your robots rules don’t block OAI-AdsBot or OAI-SearchBot on pages you do want used, since OpenAI says landing pages must not block them.
- Move tracking up a level. Put your UTM and dynamic parameters at campaign or ad group level as well as the ad URL, then watch your analytics for ChatGPT traffic landing on pages you didn’t choose.
- Send section 3 of the Ad Tools Terms to legal. One page, one question: are we comfortable owning claims we can’t preview, in languages we may not read? For regulated categories, the answer should be a campaign-level off switch.
- Get four answers from OpenAI in writing. Does every generated variation pass ad review before serving? Can we see which variations ran? Can we turn off translation or URL changes separately? Does the October 29 switch apply to campaigns bought through partners or a sales team, not just self-serve?
- Agencies, check every client account. Account defaults are per account. One forgotten client account is one client whose ads start being written by someone else next Thursday.
The pitch is less work. The terms say more responsibility. Both are true, and the toggle is set for you. Whatever you decide, decide it before OpenAI decides it for you.
