Meta AI Features Terms (Nov 20): What to Turn Off First

What changes on November 20

From November 20, 2026, Meta's Commercial Terms make the advertiser "solely responsible" for anything Meta's AI features create or change in an ad, including enhancements you never consciously turned on. The date most articles still quote, October 30, is wrong now. Meta pushed the effective date back three weeks after publishers and advertisers complained to EU and UK competition authorities.

We checked the source ourselves this morning (October 30, 2026). Meta's live Commercial Terms page carries a banner: "We're updating the Meta Commercial Terms. These updates will go into effect on November 20, 2026." The preview of the new version ends with "Effective November 20, 2026." The Self-Serve Ad Terms moved with it; a Meta spokesperson told Reuters on October 23 that "We are extending the effective date of our updated Commercial Terms and Self-Serve Ad Terms to 20 November."

So you didn't miss the deadline. You've got three weeks, and Black Friday sits one week after the new date. That's the real problem. Nobody wants to be untangling AI-generated ad variants on November 27.

What the AI Features clause actually says

Here is the full paragraph from Meta's terms preview, copied from the page on October 30, 2026:

"AI Features: Meta may use or make available products and features powered by artificial intelligence, machine learning, generative AI, or other automated products or features that help you understand and use our Meta Company Products and/or optimize, enhance or create Content. You acknowledge and agree that AI features may make mistakes and may not always produce accurate, complete, or suitable results. 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. You must not alter or remove any watermark or content-authenticating metadata generated by AI in connection with Content."

The current version of the Commercial Terms has no AI section at all. We searched the live text for it. Until now, the AI risk language lived in the separate Ad Creative Generative AI Terms Meta launched in May 2024, which were framed around tools you chose to use.

Three things in that paragraph matter more than the headline:

• "Optimize, enhance or create." That covers far more than image generation. Brightness tweaks, cropping, text rewrites and translations all fit inside "optimize" or "enhance."
• "Any resulting Content." No carve-out for features Meta switches on by default or tests on your account. If it ran, it's yours.
• It sits a few paragraphs above the indemnification clause. The same document says you'll indemnify Meta, "including reasonable legal fees and costs", for claims related to your content. An AI-edited claim that draws a complaint is your content.

There's also a line agencies should read twice. The license section says the content license applies to content "you or someone on your behalf (such as your agency that places an ad for you...)" makes available. If your agency runs the account, the AI edits made inside it land on the client's name, and the client will look at the agency.

Why this isn't a theoretical risk

Meta's AI already changes live ads. Adweek reported advertisers pulling ads over unwanted edits: added sitelinks, translated copy, AI-generated images and carousels. In June 2026, REI pulled an Instagram ad for a vendor's bicycle after an AI-altered product photo ran, and said Meta had "auto-enrolled us in an AI personalization tool." (Both cases are via PPC Land's coverage.)

The settings are built to drift on. Meta's own help page on Advantage+ creative says: "Some enhancements may be turned on by default, but you can turn them off at any time." In the Marketing API docs, adapt_to_placement is listed as "Default is opt-in." And there's an account-level switch, Test new creative features, that lets Meta apply "creative features in testing" to eligible campaigns.

Our view: treat every AI enhancement as off until a human has approved it for that client. For most of the service businesses we run ads for (clinics, law firms, home services, local B2B), the upside of an AI-generated background or a rewritten headline is a few points of CTR. The downside is an ad you can't defend saying something you didn't write. That trade made sense when the liability was fuzzy. It doesn't once the contract spells it out.

The 31 enhancements, and the 17 Meta labels as AI

On October 30, 2026, Meta's About Advantage+ creative page listed 31 enhancements. It labels 17 of them as using generative AI (or having the option to):

• Labeled AI (17): add animation, add music, add overlays, add video effects, background generation, create sticker CTA, enhance CTA, enhance media text, image generation, image touch-ups, show summaries, text generation, text improvements, translations, video subtitles, video touch-ups, visual touch-ups.
• Not labeled AI (14): add details to ad layout, add dynamic overlays, add product browsing, add product tags, add standard label, adjust brightness and contrast, dynamic description, flex media, highlight carousel card, profile end card, relevant comments, reveal details over time, show spotlights, store locations.

Don't read "not labeled AI" as "safe." The clause covers anything "automated" that optimizes or enhances content, and some unlabeled ones change what people see. Relevant comments puts a stranger's comment under your ad. Reveal details over time pulls copy from your website or app store page. Product browsing, which Meta's notice said it would switch on in existing ads on October 28, sends a tap into an in-app experience instead of your landing page. We covered that one in our product browsing opt-out guide.

How we sort them for a client:

• Off by default, every account: anything that writes or rewrites words (text generation, text improvements, enhance CTA, enhance media text, show summaries, translations, video subtitles), anything that generates imagery (image generation, background generation, add animation, add video effects, sticker CTA), and relevant comments.
• Off for regulated categories, testable elsewhere: visual and image touch-ups (they crop and expand; for before/after claims in med spa or dental ads, an expanded image is a problem), add music, add overlays.
• Usually fine to keep after a look: flex media, adjust brightness and contrast, store locations, profile end card.

Translations deserve their own warning if you advertise in Texas. The API docs describe translate_voiceover as generating a translated voiceover "that sounds similar to the original speaker," currently English to Spanish only, and note "The translated voiceover cannot be edited or regenerated." For a Dallas business whose owner records English video ads, that means an AI Spanish version of their voice making claims they never reviewed. If you serve Spanish speakers, write and record the Spanish ad yourself. It converts better anyway.

How to turn them off (Ads Manager)

Meta gives you four controls. Use all four.

Bulk edit enhancements per format

From Meta's help article on bulk editing: go to the Ads tab, check the ads, open the menu next to the Edit button and pick Advantage+ creative. Meta says the screen "shows you which Advantage+ creative enhancements are turned on for each of the ads you selected." Under Edit all selected ads, click Edit, uncheck what you don't want, then publish.

The catch is in a note: "all of the ads you select must have the same ad format." Sort by format first and run it once for single image, once for video, once for carousel. An account with mixed formats takes three or four passes, not one.

Opt out of testing new creative features

Go to Advertising settings → Creating ads → Creative features and uncheck Test new creative features. Meta notes only eligible advertisers see the option. If you see it, it's probably checked. This is the switch that lets features nobody has documented yet reach your campaigns, which is exactly what the AI Features clause makes you liable for.

Kill "Optimize text per person"

This one hides behind other toggles. Meta says that for image ads you have to turn off visual touch-ups, relevant comments, add overlays and text improvements; for video ads, video touch-ups, relevant comments and text improvements. Or toggle off All optimizations in the Advantage+ creative tab of Advanced preview. If any one of them stays on, your headline can show up as primary text next to a line you wrote for a different angle.

Check Advanced preview on top spenders

Settings tell you what's allowed. Preview shows what Meta can make. Open your ten biggest-spend ads, expand the preview, open Advanced preview → Advantage+ creative and look at every variant. Screenshot them. If a claim ever gets challenged, a dated record of what you approved is worth more than any setting.

If you launch ads through the API

Scripts are where defaults sneak back in. Meta deprecated the old standard_enhancements bundle starting with Marketing API v22.0; features are now set one by one in degrees_of_freedom_spec.creative_features_spec, each with enroll_status of OPT_IN or OPT_OUT. Leaving a key out isn't the same as opting out, so send the opt-out explicitly:

{
  "degrees_of_freedom_spec": {
    "creative_features_spec": {
      "adapt_to_placement": { "enroll_status": "OPT_OUT" },
      "image_touchups": { "enroll_status": "OPT_OUT" },
      "image_templates": { "enroll_status": "OPT_OUT" },
      "image_uncrop": { "enroll_status": "OPT_OUT" },
      "image_background_gen": { "enroll_status": "OPT_OUT" },
      "text_optimizations": { "enroll_status": "OPT_OUT" },
      "text_translation": { "enroll_status": "OPT_OUT" },
      "enhance_cta": { "enroll_status": "OPT_OUT" },
      "inline_comment": { "enroll_status": "OPT_OUT" }
    }
  }
}

Two gotchas from the docs. Music isn't in that spec at all: you opt out by sending asset_feed_spec.audios empty. And Meta silently drops features that aren't eligible for the format, so the only way to know what stuck is to GET the creative afterwards and read creative_features_spec back. If you run launches from n8n, put that read-back in the workflow and fail the run on any unexpected OPT_IN. That's the same "verify, don't trust the 200" rule we use in our n8n error handling setup.

What agencies should do before November 20

1. Run the audit above on every account, including the ones a client "mostly" runs themselves.
2. Put the approved list in writing. One paragraph in the client agreement or the launch brief naming which enhancements are allowed. When something goes wrong, "the client approved flex media and brightness, nothing else" is a conversation. "Meta did it" is not, because the terms say otherwise.
3. Re-check after Meta updates. New enhancements appear with no announcement. Put a monthly reminder on the calendar to re-open the bulk edit screen.
4. Measure the cost of turning things off. Some clients will lose a little performance. Compare backend revenue or booked appointments, not Ads Manager ROAS, before and after. If Meta's numbers and yours already disagree, our Meta Andromeda recovery guide covers how we rebuild creative testing without leaning on automation.

If you'd rather not do this account by account, it's part of what we check in a free audit. Get a Free Automation Audit and we'll go through your ad accounts, flag every AI enhancement that's on, and set up an n8n check that alerts you when a new one switches itself on. We also run ads management for teams that want it handled.

The short version

• The deadline is November 20, 2026, not October 30.
• The new clause makes you solely responsible for anything Meta's AI changes or creates, with no exception for defaults.
• Meta lists 31 enhancements, 17 labeled AI. Turn off everything that writes words or generates images unless a human approved it for that client.
• Bulk edit per format, uncheck Test new creative features, check Advanced preview, and send explicit OPT_OUT in API launches.

None of this is legal advice. If a client is in a regulated category, have their counsel read the new terms before November 20. The settings, though, you can fix today.