Field Notes · Liability

The platform rewrote your ad. Who answers for it?

OpenAI switched on default creative rewrites for ChatGPT ads with nine days' notice. Its terms do not say who carries a claim the model invents. Meta's reportedly do: you.

9 daysof notice before ChatGPT ads could be rewritten by default
A printed poster on an easel half covered by fresh paint, a roller resting against it
Approved on Monday. Repainted by Friday. Art direction: Fiction Tribe, produced with AI.
· 4 min read

The short version

  • OpenAI told advertisers on 20 October that ChatGPT would rewrite their ads by default: headlines, descriptions, translations, image sizes, even the destination URL.
  • One beta toggle covers all four behaviours. It defaults to on. No report tells you which version ran.
  • Meta's revised terms, per Adweek's reading, make advertisers "solely responsible" for AI edits, including ones Meta enables itself. PPC Land could not verify the quoted text.
  • Treat the toggle as a compliance record. Write its state into the SOW and screenshot it for every campaign.

Somebody in your business signed off an ad this month. On ChatGPT, that may no longer be the ad that runs, and nothing will tell you what replaced it.

OpenAI emailed advertisers on 20 October to say creative enhancements would be enabled by default on new and existing self-serve ChatGPT campaigns. PPC Land reported the details. The system may rewrite or add headlines and descriptions. It may translate ad text into other languages. It may resize images. It may change the destination URL to a different page on your site.

OpenAI's own wording is the part to read twice. These ads "can run automatically without requiring you to review or approve each version."

What did OpenAI change about ChatGPT ads?

It made itself the default editor of your copy, and gave you about nine days to object. The email named Friday 30 October as the start date. The help page said Thursday 29 October. OpenAI has not explained the difference.

9 days
between OpenAI's 20 October email and default rewrites starting on ChatGPT ads. PPC Land, 2026

Control is a single beta toggle. It covers all four behaviours, it is on unless you turn it off, and you cannot enable them one at a time. Want image resizing without machine-written headlines? You cannot have it. This applies to a product that now sells in 89 countries, with more than 100,000 advertisers having run campaigns.

What has OpenAI not told advertisers?

Most of what a compliance team would ask first. OpenAI has published no performance data for the feature. It has not said which model does the rewriting. It has not listed the languages. It has not explained how a replacement landing page gets chosen.

The gaps on the output side matter more. No label tells a user the copy was generated or translated. No report tells the advertiser which version ran. And the terms do not allocate liability for a claim the model invented.

Everyone who signed off a regulated claim now sits upstream of an editor they cannot see.

We have spent years in approval chains built on one assumption: a human saw the final asset. A default-on rewrite removes that assumption without removing anyone's signature.

Who is liable when a platform's AI rewrites your ad?

On Meta, reportedly, you are. On ChatGPT, nobody has said. Meta's revised Commercial Terms took effect on 30 October with a new "AI Features" section. Per Adweek's reading of the text, advertisers are "solely responsible" for what Meta's AI creates or changes in their ads, including features Meta switches on by itself. The same section concedes those features can make mistakes and produce inaccurate or unsuitable results. The earlier wording attached responsibility only when an advertiser chose to use the tools. That condition is gone.

$100
or your last twelve months of fees, whichever is greater: the cap on Meta's own liability. Meta Commercial Terms, as reported by PPC Land, 2026

Three cautions, because the reporting is thinner than the headlines. PPC Land could not verify the quoted passages against Meta's own preview. The widely repeated line that Meta shifted "100% of liability" comes from AdCellerant. It is not in the contract. And the case for leaving defaults on is soft. Meta's 3% average conversion lift comes from one experiment, run 6 to 21 August across roughly 117,700 advertisers, with the spread of results undisclosed. Its "almost 5% more conversions" figure for the AI Business Assistant compares self-selected adopters with everyone else. That is no controlled trial, and nobody should quote it as cause.

The precedent still stands. Advantage+ generates up to 150 variations by default. Defaults stay on.

What should you do before the next campaign goes live?

Stop treating this as a media setting. It is a compliance artefact. Three steps.

Write the toggle state into the SOW. On or off, per platform, per account. If the agency and the brand disagree about who decided, the document settles it.

Screenshot it for every campaign. Make the screenshot part of delivery, dated, alongside the approved copy. It takes a minute. It is the only proof of what you authorised.

Keep regulated brands opted out. Hold a standing opt-out for any account with substantiated claims until version-level reporting exists. You cannot defend a sentence you have never read.

Then keep the ledger: what ran, generated by whom, approved by whom. The deliverable here is the record. Right now nobody has one.

Questions people ask

Does ChatGPT rewrite ads automatically?

Yes, by default. OpenAI emailed advertisers on 20 October 2026 that creative enhancements would be enabled on new and existing self-serve campaigns, allowing it to rewrite headlines and descriptions, translate text, resize images and change the destination URL. One beta toggle covers all four behaviours and it defaults to on.

Who is liable when a platform's AI changes an ad?

On Meta, reportedly the advertiser: per Adweek's reading of terms effective 30 October 2026, advertisers are "solely responsible" for AI edits, including features Meta enables itself, though PPC Land could not verify the quoted text against Meta's preview. OpenAI's terms do not allocate liability for a claim its model invents.

How do I protect a regulated brand from AI ad rewrites?

Treat the setting as a compliance record. Write the toggle state into the statement of work, screenshot it for every campaign, and keep regulated accounts opted out until the platform reports which version of an ad actually ran. As of OpenAI's 20 October 2026 notice, no such version-level report exists for ChatGPT ads.

Sources

  1. OpenAI to rewrite ChatGPT ads by default as advertisers get days to opt out, PPC Land. Reports OpenAI's 20 October 2026 email to advertisers. OpenAI has published no performance data for the feature.
  2. OpenAI opens ChatGPT ads manager to all US businesses with CPC bidding, PPC Land. Background on the self-serve ChatGPT ads product.
  3. Meta's AI rewrites the ads. From October 30, advertisers carry the risk, PPC Land. Relays Adweek's reading of Meta's revised Commercial Terms. PPC Land could not verify the quoted passages against Meta's own preview.

Directed by people. Produced with AI.

Broadcast-quality film and animation at a fraction of the usual timeline, with a person deciding which frame survives.

See the films

Keep reading

Sales

Two in three buyers would rather skip the rep. Do not read them a PDF.

Gartner surveyed 646 B2B buyers and 67% said they would rather buy without a rep. The ones who still join the video call did not come to watch you scroll a PDF.

· 3 min read
67%of B2B buyers prefer a rep-free buying experience
Creative

Fifteen times the creative. The logo is still wrong.

Wyndham's five-person team claims 15 times its former creative volume. A week earlier, a brand marketer said agency work arrives with the logo wrong.

· 4 min read
15xthe creative volume Wyndham's five-person team claims, unaudited
Every note