What ChatGPT Ads' automatic rewrites mean for UK advertisers

On 29 October 2026, OpenAI will switch on creative enhancements for every new and existing ChatGPT Ads campaign that hasn’t opted out, letting its system rewrite headlines, translate copy, resize images and change landing pages. Those versions can run without anyone at the brand approving them, so UK advertisers have eight days to choose which campaigns keep the setting.

Key takeaways

• OpenAI’s help centre says creative enhancements will be enabled automatically for new and existing ChatGPT Ads campaigns starting Thursday 29 October 2026.
• Enhancements can change four things, the headline and description, the language, the image size and the destination URL, using the existing ad, its landing page and the conversation.
• OpenAI says enhanced versions can run without review or approval, and Ads Manager won’t preview every variation.
• Existing account and campaign opt-outs are kept, and a campaign setting overrides the account default, so a brand can run enhancements on some campaigns and not others.
• When OpenAI launched the feature on 16 September 2026 it was something advertisers had to opt into, and it covered text and translation only.

How OpenAI changed the setting

OpenAI announced “AI-powered text customization” on 16 September 2026, alongside Sponsored Agents and the Shopify app we covered in our post on ChatGPT Ads for UK ecommerce. Advertisers had to opt in, and it adapted existing headlines and descriptions to the conversation and translated them into the user’s language. We suggested leaving it off until a brand had read what it wrote.

The new creative enhancements page in OpenAI’s help centre widens the feature and flips the default. It adds image resizing and destination URL changes, says variations “may be prepared in advance or created in real time based on the conversation”, and states that “when enabled, these ads can run automatically without requiring you to review or approve each version.”

Which date, 29 or 30 October?

The help centre says Thursday 29 October. The email OpenAI sent advertisers, shown by Search Engine Roundtable on 20 October, says Friday 30 October and refers to self-serve campaigns, where the help page says new and existing campaigns. OpenAI hasn’t explained the gap, so set your choice before the 29th and treat accounts with a sales contact as covered unless OpenAI tells you otherwise.

What can OpenAI change in your ad?

OpenAI’s own example turns “Fresh groceries delivered” into “Groceries for quick weeknight dinners” for a conversation about weeknight cooking, which is the harmless end of the range. The page also asks advertisers to keep prices, promotions, product details and availability current on their landing pages, because the system draws on them. A sale page left live after the sale ends becomes source material for an ad.

Who answers for the rewritten ad?

In the UK, the brand does. Rule 1.8 of the CAP Code says “primary responsibility for observing the Code falls on marketers”, and rule 1.10 gives marketers primary responsibility for making sure their ads are legal. Agencies and publishers also accept an obligation under 1.8, but nothing in OpenAI’s help page moves responsibility for an enhanced headline off the advertiser whose name sits on it.

That matters most where one word changes the claim, such as a price, a “free” offer, a health benefit or a comparison with a rival. A system writing in real time from a conversation can produce a line no compliance reviewer saw, and in a regulated category that line is the brand’s problem.

What it means for paid and brand teams

Paid teams lose a clean read on their own creative. OpenAI’s reporting covers impressions, clicks, spend, click-through rate and attributed conversions at campaign, ad group and ad level, and its help page warns that “changes over time do not, on their own, establish the effect of creative enhancements.” A campaign that improves in November may owe the gain to rewritten copy, a new landing page OpenAI chose, or the season, and Ads Manager won’t separate them. The attribution tear-down playbook sets out how to keep a platform’s claims apart from your own evidence.

For brand teams the loss is the wording itself. Until now, a brand had to opt in before any ChatGPT ad ran with copy nobody at the brand or its agency had approved, and OpenAI’s own 16 September announcement said advertisers could review and edit Ads Manager’s suggested copy before using it. Our brand guardrails as code playbook turns the rules an approver applies in their head into checks, and OpenAI’s system will apply none of them unless they are already visible in the ad and the landing page.

Our view at Manual Focus is that UK brands should keep creative enhancements off at account level and switch them on for one test campaign each, where the claims are simple, the landing pages are current and a person reads a sample of the ads every week. Turn them on more widely only once a properly split test shows lower cost per conversion with enhancements on. Meta already generates creative from a brand’s own assets in Advantage+, as our guide to Meta and TikTok’s Advertising Week tools shows, but OpenAI’s version can write the copy in real time from each conversation.

What to set before 29 October

1. Set the account default to off. Do it in Ads Manager under Manage settings, and record the date and who changed it.
2. Pick one test campaign. Choose a product with no regulated claims and a stable price, and turn enhancements on at campaign level, since that setting overrides the account default.
3. Plan a fair comparison. A duplicate campaign with enhancements off is a rough guide at best, because the two compete in the same auctions and the platform can deliver them to different conversations. Treat it as directional, and before you widen the setting agree a holdout or geo-lift design with finance. Our ChatGPT Ads measurement post covers the geo-lift partners OpenAI named and the conversion setup a test needs.
4. Audit the landing pages. Remove expired offers, out-of-stock products and pages without tracking from anywhere an enhanced URL could send a click, because OpenAI says the destination may move to “a relevant page on your website”.
5. Write down your no-go claims. List the prices, comparisons and phrases your brand can’t use, and check a sample of live ads against the list each week. Our evaluation frameworks playbook shows how to score a sample consistently.

Gaps in OpenAI’s explanation

OpenAI hasn’t said whether UTM parameters on your creative URLs carry over when the system changes the destination, which affects every analytics report that relies on them. It doesn’t say whether Ads Manager will show which version of an ad ran, or let you see the rewritten text after the fact. The help page doesn’t say how translated ads are chosen for UK users, or whether a brand can limit the languages. The 29 and 30 October dates still disagree, and OpenAI hasn’t confirmed whether accounts managed by its sales team follow the same default.

Further reading

• What ChatGPT Ads on Shopify means for UK ecommerce brands, on the September launch where text customisation first appeared
• What ChatGPT’s new ad measurement tools mean for paid teams, for the conversion setup a fair test needs
• Paid channels in the AI era, on platforms taking over more of the buying and the creative
• Brand guardrails as code, for turning brand rules into checks a reviewer can run
• Attribution tear-down, for reading platform-reported results against your own evidence

Our services page explains how we help paid and brand teams set up AI advertising tests.

I run marketing for [brand], [one line on what you sell] at [website], selling
mainly in [market].

From 29 October 2026, ChatGPT Ads can change an ad's destination URL to
another "relevant page" on our site, and it reads our landing pages for
prices, promotions, product details and availability when it rewrites ads. I
want to find pages that could mislead a customer if an ad sent them there.

Attached are:
- the list of destination URLs in our ChatGPT Ads campaigns, with the
  campaign name for each
- our sitemap, or a crawl export of the site

For each campaign, look at the destination URL and the pages in the same
section of the site (same folder, category or product range). Check each page
and flag any that show:
1. an offer, discount or promotion with an end date that has passed
2. a product marked out of stock or discontinued
3. no analytics or tracking tag that you can detect

Give me a table with the columns: Campaign | URL | Issue (expired offer, out
of stock, no tracking) | What the page says | Suggested fix.

Group the rows by campaign. If you couldn't open a page, or couldn't tell
whether tracking is present, say so in the row rather than guessing.

Set this up as a routine that runs every Monday morning.

We advertise [brand] in ChatGPT Ads with creative enhancements switched on
for one test campaign, so some versions of our ads are written by the
platform and nobody at the brand approves them first.

Here is our list of claims, prices and phrases the brand can't use:
[paste list, for example banned comparisons, words like "free" or "cheapest",
health or performance claims, prices we no longer charge].

Attached is this week's sample of live ads (headline, description, landing
page and the date seen).

Check every ad against the list. Give me a table with the columns: Date seen |
Headline | Description | Landing page | Pass or fail | The words that broke
the rule | Which rule.

Then list any ad that doesn't break a rule but makes a claim our landing page
doesn't support, with the wording and the page. Don't judge tone or style
unless it breaks a listed rule. If an ad is in a language you can't read
reliably, mark it "needs a human check".

We want to test whether ChatGPT Ads' creative enhancements lower our cost per
conversion before we switch them on beyond one campaign. A duplicate campaign
with the setting off isn't a fair comparison, because the two campaigns
compete in the same auctions and can be shown in different conversations. I
need a holdout or geo-lift test plan that our finance director will accept.

Attached are:
- our ChatGPT Ads conversions and spend by week for the campaigns in scope
- the regions we sell in, with each one's share of sales

Draft a one-page test plan with these sections: Question we're answering |
Design (holdout or geo-lift, and why) | How the regions or audiences are split
| Length of the test | Budget | The result that would count as a win, and the
result that would count as no effect | Risks to the result.

Base the test length on our actual weekly conversion volumes, and say plainly
if our volumes are too low for a reliable answer in a reasonable time. List
anything the plan assumes that you couldn't check in the files.