For most of 2026, "optimizing" a ChatGPT ad meant doing it yourself — writing a few headlines, exporting an image at the right size, pointing the click somewhere sensible, and reading the auction. At the end of October that quietly changed. OpenAI turned on a setting called creative enhancements by default for new and existing self-serve campaigns, so the ad system can now rewrite your headline and description, translate the copy, resize your image, and redirect the click to a more relevant page on your own site — matching each change to the conversation the ad appears in, and doing it without asking you to approve each version. That split the job in two. One layer of optimization is now the machine's: automatic, conversation-matched, and largely invisible, running whether you watch it or not. The other layer is the one that actually decides whether the machine helps or hurts, and it is still entirely yours — the quality and on-brand consistency of the source creative you supply, the opt-in-or-out decision you make per campaign, the landing pages the system is allowed to send clicks to, and how many genuinely distinct approved variations you bring. This guide is about that second layer. It explains exactly what the automatic enhancements change and what they cost you in control, which levers survived the automation and are now where your effort pays off, why the quality of the ad is capped by the quality of the inputs no matter which posture you choose, and how to run one on-brand asset set that performs whether you leave enhancements on or opt out. The through-line: OpenAI automated the easy half of creative optimization and left you the half that matters — feeding the system creative good enough that a remix of it is still on-brand, and governed enough that a rewrite can't wander off your claims.
Until late October 2026, optimizing a ChatGPT ad was a manual job with familiar parts: you wrote a headline and a short description, exported an image that stayed legible at a small size, pointed the click at a relevant page, launched a few variations, and read the auction. Then OpenAI changed the default. It turned on a setting called creative enhancements for new and existing self-serve campaigns, and with it on the ad system itself now reshapes a running ad to fit the conversation it appears in — which means half of what used to be your optimization work is now done automatically, in the background, whether or not you are watching it.
The reporting around the rollout put the switch at the end of October 2026; OpenAI's own advertiser email and help page briefly carried slightly different dates in the days before it landed, so treat the exact day as "end of October" and confirm the current state on OpenAI's advertising help pages, since this is new and still moving. The important fact is not the date but the structural change: creative optimization inside ChatGPT Ads is now two layers. One is the machine's — automatic, conversation-matched, and largely invisible. The other is the one that decides whether the machine helps or hurts, and it is still entirely yours. This guide is about getting the second layer right, because it is the only one you can actually influence.
One toggle governs four kinds of automatic change, and it is worth being precise about each because they alter different things you may be accountable for. The system can rewrite or write new headlines and descriptions to match the conversation; translate the ad text into another language for other-language ChatGPT users; resize your image to fit the format without you re-exporting; and change the destination URL to a more relevant page on your own site. It draws on the conversation context or the user's interests together with your existing ad and landing-page content, and variations may be generated ahead of time or in real time.
Two design choices define the cost side of that convenience. First, the variations run without each one being approved by you, and Ads Manager does not preview every version — a real change from the earlier text-customization option, which produced copy you still signed off on. So an auto-rewritten headline, a machine translation, or a reassigned landing page can go live in a conversation you never saw. For a flexible direct-response campaign that is often fine; for regulated claims, exact pricing, or a disclosure you are legally on the hook for, it is a genuine exposure, because a rewrite or a translation can quietly alter the thing that has to stay exact. Second, OpenAI cautions on measurement: it says changes in results over time do not, on their own, establish the feature's effect, and it points to no report that separates an enhanced version from your original. You cannot cleanly read whether the automation helped, which is a reason to control the inputs you can read rather than trust an uplift you cannot.
Strip away the automatic layer and what remains is the layer that actually sets the ceiling. Four levers are still fully yours, and because the machine now handles the mechanical reshaping, these are where your optimization effort now pays off instead of being spent on manual resizing and rewriting.
This is the first and most consequential control, and the trap is treating it as one account-wide switch. The setting exists at both the account level (the default) and the campaign level, and the campaign setting overrides the account default — so an account-level opt-out will not switch off a campaign where the feature is already on. You have to check each campaign individually. Decide it as control versus convenience: opt out where the voice or the claims must stay exactly as written, leave it on where conversation-matching and auto-translation are a plus and the copy is forgiving. Any opt-out you had already saved before the default flipped was preserved.
The other three levers are all about what you feed the system. The source creative — the copy, the image, the brand consistency — is the raw material every automatic change starts from, so its quality propagates to every remix. The landing pages matter because the system reassigns the click among pages on your own site, which means the optimization is only as good as the set of pages you have published and allowed it to choose from; a thin site gives it nowhere useful to send a warm click. And variation depth is the lever that matters most when you opt out: with enhancements off, the only way to optimize is to bring enough genuinely distinct, approved creatives to test by hand in the auction, so a deep library is the difference between a real test and a single guess. All three are production problems, not dashboard settings.
Here is the shift most advertisers will miss. When a human did the optimizing, a strong operator could partly rescue a mediocre asset — tighten the headline, re-crop the image, salvage a weak original with craft at the editing stage. The automatic layer does not work that way. It remixes; it does not rescue. A conversation-matched rewrite of weak copy is still weak, a resized version of an off-brand image is still off-brand, and a translation of an unclear claim is an unclear claim in another language. The ceiling on everything the machine produces is set by the creative you hand it, which means the single highest-leverage thing you can do is make that input genuinely strong and genuinely on-brand — because you no longer get a second pass to fix it downstream.
Brand drift is the sharp edge of this. The faster and more automatic the remixing gets, the more a loose or inconsistent source creative can wander — a headline rewritten away from your positioning, a translated line that no longer matches how you describe yourself, an image resized in a way that loses the layout your brand depends on. The defense is not to fight the automation case by case (you cannot — it runs without your approval) but to make the source so tightly governed that even a remix of it stays inside your voice and your claims. Optimization, in other words, became an upstream discipline: fix the inputs so hard that whatever the machine does with them is still something you would have shipped. The broader strategic case for why consistency beats volume is in AI ad creative iteration; here it is specifically the thing that makes the automatic layer safe.
The practical answer is not to build two different creative strategies for "enhancements on" and "enhancements off." It is to build one deep, governed, on-brand asset set that performs either way. If you opt out, you need enough distinct approved variations to run a real test against an intent, so depth is the requirement. If you leave enhancements on, the rewritten headline and resized image are still built from your source, so on-brand strength is the requirement. Both paths reward the same thing — a library of creative that is both deep and tightly consistent — which is why the posture decision and the production decision can be separated: produce the strong governed set first, then choose per campaign whether to let the machine remix it or test it by hand.
Pair that with the few things the automation does not touch. It optimizes the ad; it does not write your landing pages, run your prompt-tracking panel, or earn you the organic citation that sits above the paid unit in the same answer. That organic slot is a separate, more-trusted, unbuyable surface, and running it alongside the ad is what warms the click before it lands — the full two-slot workflow is laid out in how to build a ChatGPT ads and GEO content strategy, and the state of the early performance data, given that OpenAI publishes no benchmarks, in ChatGPT ads performance. The feature mechanics themselves are catalogued in ChatGPT Ads creative enhancements.
Because the optimization moved upstream to the inputs, the question that decides your ChatGPT ad results is no longer "how do I tune the ad" — the machine does that — but "how do I produce a deep set of creative that is strong enough to survive a remix and governed enough that a rewrite can't wander off my claims." That is a generation problem, and it is exactly what Kompozy is built for. It is a full AI content generation and multi-platform publishing engine, not an ad tool, and brand-exactness is its design point: a Persona Brief fixes the voice, claims, and banned words that every generation must hold to, Gemini face-lock keeps a persona's face identical across images, and HyperFrames renders pixel-exact layout and wordmark placement. That governance is precisely the thing the automatic layer cannot guarantee on its own — and the thing that makes leaving enhancements on survivable, because a remix of a brief-governed asset starts on-brand.
The workflow is concrete and serves both postures. From one brief, generate a deep batch of on-brand units — Photo Posts and Quote Graphics, Infographic Photos, brand-exact Carousel Posts, and short-form video like Persona Shorts and Marketing Shorts — so you have genuinely distinct variations rather than one asset and a hope. Opt a campaign out and that depth becomes the test set your own CPC and conversion data ranks; leave it on and the same assets give the system a strong, consistent source to remix. Either way, every asset you made also publishes as organic posts across eight social platforms plus blog and email through Autopilot and a per-post review gate — which is what builds the citable footprint that earns a ChatGPT mention in the answer above your ad, the one surface no optimization toggle controls. The honest boundary: Kompozy produces and publishes the creative both postures run on; it does not bid in OpenAI's Ads Manager, flip the enhancements toggle, or read the auction for you. It supplies the governed input layer that the automation, by its own design, can only ever reshape.
ChatGPT Ads creative optimization stopped being one job at the end of October 2026. OpenAI's default-on creative enhancements took over the mechanical half — rewriting headlines, translating copy, resizing images, and redirecting clicks to match each conversation, without asking you to approve each version. That leaves you the half that decides the outcome: the quality and brand-consistency of the source creative, the opt-in-or-out posture you set per campaign, and the landing pages and variation depth you bring. Because the machine remixes rather than rescues, input quality is now the ceiling on everything, so the winning move is to feed the system a deep, tightly governed, on-brand asset set — strong enough that a remix of it is still something you would have shipped, and consistent enough that a rewrite can't wander off your claims.
It is now two jobs, not one. OpenAI's creative enhancements setting — on by default since late October 2026 — automatically optimizes the running ad to fit each conversation: it can rewrite the headline and description, translate the copy, resize the image, and change the click destination to another page on your site. That is the machine's half. Your half is the inputs it works from — the source creative, the opt-in-or-out decision, the allowed landing pages, and the depth of approved variations — which is where the real optimization work moved.
It took over the mechanical half. With creative enhancements on, the system reshapes a running ad to match the conversation using your existing creative and landing-page content, and it can do so without you approving or previewing each version. But it only ever remixes what you supplied — it generates no net-new ad — so the parts that decide whether the result is good are still yours: the quality and brand-consistency of the source, and whether you let the feature run at all. The optimization did not disappear; it moved upstream to the inputs.
Treat it as a control-versus-convenience call made per campaign, not once for the account. For flexible direct-response campaigns where reaching other-language users or auto-resizing is a plus, leaving it on is reasonable. For regulated industries, precise pricing, or a tightly governed voice, opt out — because variations run without per-version approval and Ads Manager previews no version, so an auto-rewritten headline or translated line can alter a claim or disclosure you are accountable for. The campaign setting overrides the account default, so check each campaign individually.
Only as good as what you feed it. The enhancements reshape your source creative to fit a conversation, so a weak or off-brand original produces a weak or off-brand remix — the ceiling is set by the inputs. OpenAI also cautions on measurement: it says changes in results over time do not, on their own, establish the feature's effect, and Ads Manager shows no report that isolates an enhanced version from your original. So you cannot assume the automation lifted performance; you can only make sure the creative it starts from is strong.
Four things, and they are where your effort now pays off. The source creative you upload — its copy, image quality, and brand consistency — which caps everything downstream. The opt-in-or-out posture per campaign. The landing pages on your site the system is allowed to redirect clicks to, since it reassigns destinations among your own pages. And the depth of genuinely distinct approved variations you bring, which matters most when you opt out and have to test by hand. The lever that survived the automation is input quality.
ChatGPT Ads creative optimization split into two layers in late October 2026, when OpenAI turned on creative enhancements by default. One layer is now automatic: the system rewrites headlines, translates copy, resizes images, and redirects clicks to match each conversation, without per-version approval. The other layer is still yours and decides the outcome — the quality and brand-consistency of the source creative, the opt-in-or-out decision per campaign, and the landing pages you allow. The lever that survived is input quality.
Get started → · ← All guides · Compare Kompozy vs other tools