Imagine approving an ad, then finding out that the version people actually saw was not quite the one you approved. A headline was translated. A sitelink appeared. A generated image replaced part of a photo from a professional shoot. Nothing in your change log explains it, because you did not make the change. Meta’s software did.
That gap between the ad you approved and the ad that ran is now the central issue in Meta’s revised Commercial Terms. The new text adds an “AI Features” section that, on Adweek’s reading, makes advertisers solely responsible for whatever Meta’s AI features create or change in their ads, including features Meta itself switches on. The terms also say those features may make mistakes. Put those two sentences together and the practical meaning is plain: if the machine edits your ad, the public, the platform, and possibly a regulator will still treat it as your ad.
There is one important correction to the headlines you may have seen. These terms were originally due to take effect on October 30, 2026. They did not. Meta first delayed the date to November 20, then paused the rollout on October 28 pending consultation with industry groups. The existing terms remain in place until then, and no new end date has been given. That pause is good news, but it is not a reprieve from the work. It is a window to audit your account while the wording is still, in Meta’s own framing, under discussion. Use it.
The short version
What changed: Meta’s revised Commercial Terms add an “AI Features” section. It says advertisers are solely responsible for AI features and any resulting content, including judging whether that content is suitable for business use.
What it covers, on Adweek’s reading: AI edits to ads, including features Meta switches on by default, not only tools an advertiser deliberately chooses.
What is not yet in force: The revised terms. The October 30 date moved to November 20, and on October 28 Meta paused the rollout while it consults industry groups. The consultation could extend beyond November 20.
What is already happening: Agencies report Meta’s tools adding sitelinks, translating copy, and generating images and carousels from brand photos inside live campaigns, sometimes without the advertiser asking for it.
What to do now: Audit every live ad against its approved version, switch off enhancements you cannot stand behind, protect AI watermarks and metadata in your export pipeline, and put a named person in charge of checking for machine-made changes before Black Friday.
What the new wording actually says
The change is small in length and large in consequence. Previously, advertisers carried responsibility for AI output when they chose to use Meta’s ad creative AI tools. Adweek quotes the earlier wording as giving advertisers the choice to use those tools “in your sole discretion.” The new section drops that condition.
The operative sentence in the preview, as quoted in secondary coverage of the terms, says: “You are solely responsible for and assume any and all risks associated with such AI features and any resulting Content, including evaluating the suitability of any Content for business or commercial purposes.”
Read it alongside the sentence that comes before it, in which Meta acknowledges that AI features “may make mistakes and may not always produce accurate, complete, or suitable results.” The contract, in other words, anticipates errors and assigns them to the advertiser in advance.
A second obligation is easy to breach by accident. The terms say advertisers must not alter or remove any watermark or content-authenticating metadata generated by AI in connection with content. Many normal production steps strip metadata without anyone deciding to strip it: an export preset, a resize, a compression pass, or a re-upload through a digital asset manager. Trace one AI-generated asset from generation to live ad and check whether its credentials survive each step. If they do not, fix the pipeline before you argue about the policy.
There is also the question Meta has not directly answered. A Meta spokesperson told Adweek that advertisers can turn off individual Advantage+ features in Ads Manager at any time. The spokesperson did not directly say whether an advertiser is liable for a feature it never switched on. That unanswered question is the whole story for many advertisers, because the reported cases are not about tools people chose. They are about tools that arrived through an account update or a default setting.
This article is a marketing operations guide, not legal advice. If your ads make regulated claims, carry prices, or use licensed assets, have your legal team read the preview text itself, not just coverage of it.
How the dates moved, and why the confusion matters
The timeline moved fast enough that much of the coverage now circulating is out of date. Here is the sequence, as reported by PPC Land, Good Morning Co, and Adweek’s October 22 report:
In mid-October, Meta published a preview of revised Commercial Terms and Self-Serve Ad Terms. The preview included a crawling clause for business websites that drew publisher complaints, and the new AI Features section for advertisers.
On October 17, Meta removed the word “train” and related training language from the crawling clause after criticism. Coverage published before that date still quotes wording that is no longer in the preview. If someone on your team forwards an older article claiming Meta will train its AI on your website under these terms, check the preview first. That claim describes an earlier draft.
On October 22, Adweek reported the AI Features section and the advertiser reaction to it.
On October 23, Meta extended the effective date from October 30 to November 20, saying it needed more time to explain to partners what is changing and what is not. The wording, according to Good Morning Co’s October 29 update, did not change. Advertisers got three more weeks, not a different contract.
On October 28, two things happened on the same day. An Oslo court temporarily blocked Meta’s new terms on an application from the Norwegian Media Businesses’ Association, and Meta announced a pause to the rollout pending consultation with industry groups. Meta’s vice president of communications, Andy Stone, said the company did not plan to make policy changes without a full consultation, that the consultation could extend beyond November 20, and that the existing terms would remain in place until then. No end date was given, and no change to the text was announced.
On October 30, the UK Competition and Markets Authority opened a formal investigation into Meta’s commercial terms for businesses, under Chapter II of the Competition Act 1998. Its concern is whether terms that let Meta access and use content from business and publisher websites, without those businesses getting a choice on how it is used, amount to unfair trading conditions imposed by a dominant firm. The CMA has made no finding. Its indicative timetable runs initial information gathering from October 2026 to January 2027, and it states plainly that no assumption should be made about whether competition law has been broken.
The lesson for marketers is not just about Meta. It is about how fast contract news decays. A deadline that was correct on October 22 was wrong by October 23, and a November 20 date that looked firm on October 29 was overtaken by a pause on October 28. Before you brief a client or a leadership team, check the live preview and the latest regulator statement, and date-stamp your own summary so nobody acts on last week’s version.
What is already happening in live campaigns
The contract debate would matter less if the tools only changed ads when asked. According to agencies interviewed by Adweek, that is not how it works in practice.
Bookshop.org, the online bookseller, said in a post on Threads that an automated account update opted it into a generative AI feature, which added an AI-generated illustration to an ad its staff had designed, without its knowledge. Adweek reports that Bookshop.org has since turned off all of Meta’s AI ad features.
Exverus, a digital agency, told Adweek that Meta’s AI is adding sitelinks, translating copy, and producing AI images and carousels from a brand’s own photos, sometimes on ads where advertisers did not switch the feature on. The agency said it has had to pull live ads to remove changes that got through several rounds of quality checks. Its paid search and social specialist, Ryan Adams, described the settings problem directly: Meta auto-applies the enhancements, and many options sit in hidden or collapsed menus, so if every one is not manually unchecked, the enhancements still apply.
There is an earlier precedent worth remembering. In June 2026, REI pulled an Instagram ad for a Van Rysel bicycle after an AI-altered image added a second set of handlebars, illegible frame text, and a missing crank arm to a photo from a professional shoot. REI said on June 22, 2026, that Meta had enrolled it in an AI personalization tool without its knowledge. PPC Land reported at the time that Meta had not publicly addressed the enrollment mechanism.
None of this means every AI enhancement is harmful. Many advertisers use Advantage+ creative features deliberately and report good results. The risk is narrower and more practical: an enhancement you never reviewed can change a claim, a price, a face, a product detail, or a brand asset, and the audience will assume you meant it. As Emarketer senior analyst Asa Hiken told Adweek, consumers are unaware of Meta’s ad policies and are likely to assume the brand willingly used the AI version.
What did not change, and what is still disputed
Three clarifications will save your team from chasing the wrong problem.
First, the training clause is not the clause it was in mid-October. The word “train” was removed from the preview on October 17, and the permitted purposes in the current preview are to provide, develop, improve, research, and maintain advertising and business products. The crawling clause is still worth a deliberate decision, especially for publishers and businesses whose content is itself the product, but argue about the clause as it stands, not the draft that was withdrawn.
Second, the testing clause is not new. The Self-Serve Ad Terms already say Meta runs tests on its self-serve interfaces from time to time, which may affect performance, and that permission has been in the live terms since August 2025. It appears in the preview unchanged. Unexplained performance movement may still be a test you were never told about, but nothing about it changes on November 20.
Third, the liability question for features an advertiser never enabled is reported, not settled. Adweek’s reading is that the AI Features section reaches features Meta switches on itself. Meta’s position, as stated to Adweek, is that the update folds existing generative AI ad terms into the main terms and that each feature can be switched off. Those two accounts can both be true at once: a toggle can exist and still be easy to miss, and a contract can assign risk regardless of whether the toggle was found. Until Meta answers the never-enabled question directly, or the final text says so explicitly, treat it as an open risk, not as a proven fact in either direction.
The same caution applies to the CMA case. An investigation is not a verdict, and the pause is not a withdrawal. The revised terms could return in their current form, in amended form, or after a longer consultation. Planning for the current wording, while noting that it is paused, is the honest middle course.
Why defaults are the real story
Strip away the legal language and this is a story about defaults. Meta has been moving toward default-on creative automation for more than a year. In October 2025, Advantage+ creative enhancements were activated by default, adjusting backgrounds, brightness, contrast, and format, and generating variations from advertiser assets. Each default widens the set of ads that carry machine-made changes. The AI Features section, on Adweek’s reading, assigns the outcome of all of them to the buyer.
Defaults are powerful because they survive inattention. A setting chosen once, at account level, does not necessarily govern every ad. Enhancements are set per ad, so a duplicated ad, an imported ad, or a new ad created from a template can start from Meta’s defaults again. An agency that switches features off “across every campaign,” as AdCellerant’s director of operations, programmatic, Kayla Loecker, told Adweek her firm does, still has to check the next campaign, the next duplicate, and the next new format.
This pattern is not confined to Meta. Two days before the original October 30 date, Meta switched on product browsing for existing Facebook and Instagram ads unless advertisers opted out, sending some people who tap an ad to an in-app product page rather than the advertiser’s website. OpenAI, meanwhile, told advertisers it would enable creative enhancements by default for new and existing self-serve ChatGPT campaigns around October 29 to 30, allowing its system to rewrite headlines and descriptions, resize images, change destination URLs, and translate ad text. The platforms are converging on the same model: the system improves the ad, the advertiser owns the result, and opting out is a manual, per-ad job.
That model can still be worth using. Automation at Meta’s scale finds combinations no human team would test by hand, and many advertisers accept the trade. The mistake is accepting it by default, without deciding which kinds of change are acceptable for your brand. A background brightening is not the same as a rewritten claim. A translated headline is not the same as a generated product image. Decide where your line sits, then set the toggles to match it, rather than letting the platform’s line become yours.
The audit to run before November 20
Do this before the holiday peak, while the pause gives you room to fix things without a live deadline. It takes an afternoon for a small account and a week for a large one, and it is the same audit whether the terms return on November 20, later, or in amended form.
1. List every live ad and its approved version. Export your live ads, then compare each one against the creative your team actually approved. Look specifically for added sitelinks, translated copy, altered backgrounds, generated images or carousels, and changed headlines or descriptions. Screenshot anything that differs. Your change log will not show machine-made changes, so the screenshot is your record.
2. Open the enhancement settings on each ad, not just the account. Check collapsed and hidden menus. If an enhancement is on and you cannot explain why, switch it off and note the decision. When you duplicate an ad or create a new one from it, re-check the settings before it goes live. Build this check into your launch checklist so it survives staff changes.
3. Sort enhancements by risk, not by format. Put any change to a claim, a price, a regulated statement, a person’s face, or a licensed asset in the “never without human sign-off” group. Put background, brightness, and format adjustments in a lower-risk group, and only if your brand guidelines allow them. If a tool can generate a new image from your photos, decide in advance whether that is allowed at all.
4. Protect watermarks and metadata. Follow one AI-generated asset from generation to live ad. Check whether its watermark and content credentials survive export, resizing, compression, and re-upload. If a step strips them, change the step or the preset. This is now a contract obligation in the preview text, not just good practice.
5. Brief legal, finance, and client teams in one paragraph. Tell them acceptance is passive: continued use after the terms take effect counts as acceptance, with no signature and no checkbox. Agencies accept on behalf of the accounts they operate, so clients should hear it from their agency before they read it elsewhere. Keep the note short, factual, and dated, and link to the preview rather than to coverage of it.
6. Make the robots.txt decision separately. The crawling clause is a different question from ad liability, and it deserves its own decision. Blocking every Meta user agent can break link previews and degrade catalog surfaces. Decide per agent, with legal input, rather than in one blanket rule. Publishers and course businesses, whose editorial archive has standalone value, should think hardest about this one.
7. Keep a log of what the platform changed. Record the date, the ad, the change, and the fix. If performance moves and your change log shows nothing, that log becomes the first place to look, and the first evidence if a client or a regulator asks what happened.
8. Set a review date, not just a deadline. The terms are paused, not withdrawn, and the CMA investigation runs into January 2027. Put a recurring check in the calendar for the preview text and the regulator case page, so a quiet return of the terms does not catch you in the middle of a campaign.
What to do if an AI edit slips through
Pause the ad first, then investigate. Do not debate the wording while a wrong claim is still running. Screenshot the live version, note the time, and check whether the change came from an enhancement setting, an account update, or a duplicated ad that reset to defaults. Fix the setting, not just the single ad, or the next ad will repeat the error. Tell the client in plain language what changed, what you did, and what you changed in the process to stop it happening again. If the edit touched a regulated claim, a price, or a licensed asset, involve legal before you republish.
It is also worth telling Meta through your normal support or representative channel. One report will not change a contract, but a pattern of reports, like the agency accounts in Adweek’s coverage, is how these issues become visible. The Bookshop.org and REI cases became stories because someone noticed, documented, and spoke up.
The takeaway for marketers
The platforms are no longer just places where you publish ads. They are editors of your ads, and they are increasingly editing by default. The contract is catching up with that reality by assigning the result to the advertiser. Whether you think that is fair is a separate question from what to do about it, and only the second question is fully in your control.
Do not wait for the final terms to decide your standard. Decide now which machine-made changes your brand will accept, switch off the rest, and put a human check between the machine and the audience for anything that changes meaning. That check is the real product here: not a toggle, not a legal paragraph, but a habit of looking at the ad people actually see, not just the ad you approved.
The pause until at least November 20 is a gift of time. Spend it on the audit, not on the argument. When the terms return, in whatever form, your account should already be in the state you would want it to be in if they had taken effect on October 30.
Sources
- PPC Land, “Meta makes advertisers solely responsible for its AI ad edits from Oct 30,” reporting Adweek’s October 22, 2026 report by Trishla Ostwal: https://ppc.land/meta-makes-advertisers-solely-responsible-for-its-ai-ad-edits-from-oct-30/
- Good Morning Co, “Meta Ads Terms Update October 2026: A Checklist,” last updated October 29, 2026 (effective date extended to November 20, wording unchanged): https://goodmorningco.com/blog/meta-ads-terms-update-october-2026
- PPC Land, “Meta faces UK probe over planned terms to use publishers’ content for AI” (CMA investigation opened October 30, 2026; October 28 pause; Oslo court order): https://ppc.land/meta-faces-uk-probe-over-planned-terms-to-use-publishers-content-for-ai/
- Relevant Audience, “Meta AI Features terms: advertisers liable from 30 Oct,” summarizing Adweek’s October 22 report: https://www.relevantaudience.com/meta/meta-commercial-terms-ai-features-advertiser-liability-october-30/
- Seafoam Media, “November 2026 Marketing News: Trends & Insights” (October 30 terms, October 28 product browsing, and the gap over features never enabled): https://seafoammedia.com/thinking/brief/november-2026-marketing-news-trends-insights/
Related reading: