This is an ADOTAT investigation. We take what a company says about itself and hold it up against its own paperwork. We put the questions in writing, we print the answers, and we print it when there aren't any. The top of the piece is free and built to be forwarded. The deeper reporting sits behind a free email signup. The field kit and the questions for your next platform call are for ADOTAT+ members. We run ads. We don't sell editorial. This week: two platforms that will rewrite your ads this week, and the contracts they rewrote first.

OpenAI sent its advertisers a friendly little email on October 20. The subject was a feature called creative enhancements, and the pitch was the pitch every platform has made since the first algorithm learned to say "optimize": more relevant ads, better performance, less work for you.

Then came the sentence doing the actual work. The new versions of your ads, OpenAI wrote, can run automatically "without requiring you to review or approve each version."

Here is what that means in English. Starting at the end of this month, unless you go find a toggle and turn it off, OpenAI can rewrite your headline, rewrite your description, translate your copy into a language your legal team does not read, resize your image, and send the click to a different page on your website. It can do that ahead of time or on the fly, tailored to whatever the person in the chat happens to be talking about. And you do not have to see it first.

Meta already flipped its own switch. Since October 28, ads already running on Facebook and Instagram have had product browsing turned on by default, which means a tap on your ad can open a Meta-hosted page with product info and personalized suggestions before anyone reaches your site. Tomorrow, October 30, Meta's revised Commercial Terms take effect.

Each of these, on its own, is a product update. The kind of thing that gets two paragraphs and a "what marketers need to know" listicle.

Put them next to the paperwork and they turn into something else. In the weeks before both switches, both companies rewrote their contracts so that the ad the machine writes is legally yours.

The approval step went away. The liability stayed exactly where it was. On you.

Start with OpenAI, because OpenAI left the better paper trail.

July 31. OpenAI publishes its Ad Tools Terms, the contract that governs its AI creative tools. The deal is simple and, frankly, fair. The machine drafts. You decide. You are responsible for the generated creatives that you approve or use, and the terms say you are "responsible for reviewing Generated Creatives before use." Liability follows approval. You signed off, you own it.

August 24. A new version of the same terms appears. This one says the creative tools may automatically apply generated creatives to your campaign. It makes you responsible for every generated creative you approve or use, "or that is applied to your Campaign." It extends your duty to indemnify OpenAI to those creatives too. It adds that OpenAI no longer promises the output will even match what you meant to say. And the sentence about reviewing creatives before use? Gone.

Read the two versions side by side and the edit is surgical. The review step was taken out of the contract at the same time the product was rebuilt to skip it. What stayed was the part where it is your problem.

October 15. OpenAI updates its main Advertising Terms. You are now solely responsible for each campaign and for any AI-generated output delivered through it, and the terms spell out that this covers any optional tool or feature "enabled by default and remains enabled for your Campaign."

October 20. The email goes out. The feature is enabled by default.

That is not a coincidence of timing. That is a sequence. First you make the default your customer's responsibility. Then you turn on the default.

One more detail, because it is the kind of thing that should not happen at a company this size. On Monday, OpenAI's own terms pages did not agree with each other. The U.S. English Ad Tools Terms page was still dated July 31 and still carried the review-before-use sentence. The UK English page was dated August 24 and did not. Which version governs a U.S. advertiser at the end of this month is not a trivia question. It is the whole question.

Meta's version is shorter and blunter.

Under Meta's earlier generative AI ad terms, as Adweek reported last week, advertisers already carried responsibility for AI output, but the clause was built on choice. You could use Meta's ad creative AI at your discretion, and if you did, the output was yours.

The revised Commercial Terms, effective October 30, add a new section called AI Features. According to Adweek's reading, it makes advertisers solely responsible for whatever Meta's AI features create or change in their ads, including features Meta switches on itself, and puts the risk on the advertiser while conceding, in Meta's own words, that its AI "may make mistakes."

There is no signature and no checkbox. Under Meta's terms, continuing to run ads after the update is how you agree to it. If you are running ads on Halloween, you said yes.

Meta told Adweek the update consolidates its generative AI ad terms into its main ad terms, and that advertisers can turn off individual Advantage+ features in Ads Manager at any time. It did not directly answer whether advertisers are liable for AI features they never switched on.

So, the two companies' positions, one sentence each:

OpenAI: We may write it. You are responsible for it. You did not have to see it.

Meta: Our AI may make mistakes. The mistakes are yours. We turned it on.

If you are trying to work out when all of this happens, welcome to the club. The platforms are not sure either.

October 28: Meta product browsing. The default flipped for ads already running. Meta's notice carries no date of its own; LinkedIn timestamps and a third-party change log put its arrival around October 14. It landed about a month before Black Friday, November 27.

OpenAI creative enhancements: pick a day. The October 20 email says Friday, October 30. OpenAI's help page said Thursday, October 29 when it was checked on October 21. Since at least October 23, the same help page has said "Thursday, October 31, 2026."

October 31, 2026 is a Saturday.

The company whose product can now translate your ad into other languages without asking could not, in three tries, line up a weekday with a date on its own help page. That is funny. It is also the page advertisers are told to rely on for the deadline that decides whether a machine starts rewriting their copy.

October 30: Meta's new Commercial Terms take effect.

The only safe reading of OpenAI's dates is the earliest one, and the earliest one is today. If you want out, turn it off now, and screenshot the setting when you do.

Strip out the marketing language and here is the inventory.

OpenAI, headlines and descriptions. New or rewritten text built to connect your product to the conversation or to the user's interests.

OpenAI, language. Your ad text can be translated into another language. ChatGPT ads now sell in 89 countries.

OpenAI, images. Resized to fit the format. OpenAI does not say whether resizing includes cropping, which matters a great deal if your logo or your legal line lives near the edge.

OpenAI, the destination. The URL can be "optimized" to send people to another relevant page on your website, chosen by OpenAI. Neither the email nor the help page explains how that page is picked or how to keep one off the list.

OpenAI, the review. Variants can be prepared in advance or generated in real time, and OpenAI's help page says advertisers do not get a preview of every variation.

Meta, the click. A tap on an existing ad may open an in-app page with product info and personalized suggestions, pulled from sources Meta's notice does not specify, before the person reaches your site. Your landing page, your consent banner, your promo test and your analytics all now sit one step further back.

Each of those used to be a brand's decision. Every one of them can now be made by the platform, and every one of them, by contract, is still the brand's decision.

Everything below this line is for subscribers. Sign up free with your email to keep reading: the contract clauses that cap what the platforms owe you, the brands already pulling ads to undo Meta's edits, and who in the chain ends up holding the bag when a machine-written ad gets something wrong.