Bookshop.org, the online bookseller, discovered that a Meta account update had opted it into a generative-AI feature. The feature added an AI-generated illustration to a human-made ad, without anyone at Bookshop requesting it. The company's response on Threads was blunt: the ad was "enhanced (or ruined, really)" without its knowledge. Bookshop.org said it then turned off all of Meta's AI ad features.
That's one bookseller. The pattern is wider, and the contractual ground underneath it is shifting fast.
What's Actually Happening in Ads Manager
Digital agency Exverus told Adweek it has seen Meta add sitelinks, translate ad copy, and generate images or carousels from brands' own photos on live campaigns. Some changes appeared on ads where the advertiser never switched the feature on. Even after several rounds of QA, Exverus said it had to pull live ads to strip out changes that made it through its checks.
Ryan Adams, Exverus's paid search and social specialist, pointed to a specific operational problem: "Meta auto-applies the enhancements, and many of the options are tucked away in hidden or collapsed menus. If every one of those isn't manually unchecked, the enhancements still apply."
For B2B SaaS teams running regulated messaging or legally reviewed copy, a translated headline, an AI-generated product image, or an auto-added sitelink can break compliance, misrepresent capabilities, or contradict positioning that took weeks to approve.
The Liability Clause Nobody Asked For
Meta's revised Commercial Terms add a new "AI Features" section. The language, as reported by Adweek, states that businesses are "solely responsible" for risks associated with Meta AI features and any resulting content, including output that may be "inaccurate, incomplete, misleading, offensive, or inappropriate." All risks sit with the advertiser.
The previous terms placed liability on advertisers too, but only when they chose to use Meta's AI tools. The new language covers features Meta enables, whether or not the advertiser opted in. When Adweek asked whether advertisers would bear responsibility for AI features they hadn't switched on themselves, Meta did not directly answer. A spokesperson said advertisers can turn off individual Advantage+ features in Ads Manager at any time.
Meta postponed the effective date of the updated terms from October 30 to November 20, 2026, citing a need for more time to explain changes to partners and stakeholders, according to Reuters.
Why This Breaks Your Experimentation Loop
If the platform can silently alter creative after your team's approval, you lose the ability to isolate what you're testing. You set up an A/B test on headline copy. Meta adds a sitelink or swaps in a generated carousel. Your conversion data now reflects a variable you didn't control and may not even know changed. That's not a measurement problem you can solve with better attribution modeling. It's a governance gap.
An eMarketer analyst, cited by Adweek, flagged a related reputational risk: consumers see the brand's ad, not Meta's workflow. If AI-generated content looks off or inaccurate, the audience attributes it to the brand. For B2B companies selling to procurement teams and security-conscious buyers, that kind of semantic drift can erode trust in ways that don't show up in your CPA.
The Ops Response: What to Do Before November 20
Audit every campaign's Advantage+ settings. Go ad-by-ad, not campaign-level. Exverus's experience confirms that controls can sit in collapsed menus. Document what's on and what's off. Screenshot it. If your agency manages the account, require them to send you the receipts.
Build a post-launch QA layer. Pre-launch approval isn't enough anymore. You need a rendered-ad check after the campaign goes live, comparing what's actually serving against what was approved. If you're running regulated copy (think: SOC 2 claims, pricing language, feature statements), this becomes a compliance requirement.
Get your legal team on the new terms. The "AI Features" clause is a contractual risk-allocation provision, not a court ruling, and enforceability will depend on facts and applicable law. But the language is broad enough that your legal and compliance teams should review it before November 20.
Log what actually ran. If you ever need an audit trail, you need proof of what creative and copy actually served across placements. Meta's reporting may not give you that granularity. Consider third-party ad verification or manual spot checks on a schedule.
Decide your stop-loss. If an AI enhancement slips through and breaks your messaging, how fast can you kill the ad? Who owns that decision? Set the SLA now.
Turning off every Advantage+ feature is the safest move for creative control. But Meta's own data (from a January 2026 post) attributed a 3.5% lift in Facebook ad clicks and more than a 1% gain in Instagram conversions in Q4 2025 to a set of improvements. Those numbers are aggregate, not broken out by feature, and they don't tell you anything about the impact of unwanted edits specifically. Turning everything off may cost you some performance lift. Leaving everything on costs you control, auditability, and potentially your compliance posture.
That's the real calculus. Your CFO, your legal team, and your RevOps lead all need to weigh in on it, not just your media buyer. Contract language is now part of the media plan. The sooner your ops team treats it that way, the less likely you are to find out about an AI edit from a customer screenshot instead of your own QA process.
