Meta Adds, Then Removes, an AI-Crawling Clause From Its Business Terms

Meta added a clause to its business terms this month that three outside legal experts read as permission to crawl advertisers' and publishers' websites and train its AI on the content. Two days after that reading reached print, the company removed it.

Related reading: Meta's Oversight Board Orders Deepfakes Removed and Calls Its Safeguards Inadequate · Meta One Puts AI and Creator Tools Behind a Monthly Subscription.

The wording appeared in an update to Meta's Commercial Terms, the agreement that governs anyone using its products for business. Nieman Lab reported the change on 15 October 2026. The new version was dated to take effect on 30 October, replacing one from 6 March 2024, according to PPC Land.

The clause sat in Section 1. It authorised Meta to use “automated software (such as web scrapers, bots, robots, spiders, crawlers, and user agents)” to access, index, cache, collect, retrieve or crawl the domains, web pages or URLs “associated with” a business's products or services, and to create “test credentials” and pull updates periodically. It added that Meta “may use any content from such domains, web pages, or URLs to provide, develop, improve, research, train, and maintain the integrity of, the Meta Company Products, services, and features.”

The terms reach further than they sound. They apply to Facebook Pages, Messenger, Instagram profiles, WhatsApp channels and ad accounts, which covers the business presence of nearly every news publisher. Accepting the update carried the permission. There was no box to uncheck, and the document offered one exit: “If you do not agree to the updated terms, please stop all access or use of our Products for business or commercial purposes.”

Consent by continuing to use the platform

The mechanism is the part that drew attention. A clause in a document most business users never read can hand over rights over property those users never meant to give up. Continue using Pages, Messenger, Instagram, WhatsApp or an ad account after 30 October, and the terms applied.

Three legal experts who reviewed the language for Nieman Lab said it could be read as permission to scrape sites associated with business accounts and to train Meta's AI on what it found. Matt Topic of Loevy & Loevy called it “a remarkable provision to bury in terms of service”. James Grimmelmann of Cornell Law told Nieman Lab that on the broadest reading it would let Meta build “a comprehensive archive” of that content. Benjamin Sobel of the University of Wisconsin likened the tie to an antitrust “tying” arrangement, where access to one product is used to secure terms over another.

The absence of an opt-out is what turned the language from routine to contentious. The only remedy the terms offered was to stop using the products for business. That is not a realistic option for an organisation that runs a Facebook Page and an ad account to reach customers, so the choice sat between accepting the terms and leaving the platform.

One phrase deserves its own label. “Test credentials” is Meta's own wording. Nieman Lab's experts said it could allow Meta to sign up to a site and bypass a registration wall, and Meta declined to answer that question when asked. It is not established that the company planned to defeat paywalls, and the clause should not be described that way.

Meta's denial, then a deletion

Meta pushed back quickly. Vice-president of communications Andy Stone and spokesperson Daniel Roberts said there was “no change to how Meta uses web crawlers; any suggestion to the contrary is false,” The Verge reported. The company said the update consolidated crawling terms already scattered across other agreements, including its generative-AI-in-ads terms, and that content owners could still rely on robots.txt.

That explanation did not close the matter. On 17 October 2026 Meta removed the disputed clause from the commercial terms page. The Verge reported that the sentence about training no longer appeared. TechCrunch reported the removal together with a fuller company statement: “We've heard you and want to clarify: the crawling terms we consolidated only applied to our ads and business products, and we've clarified that in the terms themselves. It does not change how Meta uses web crawlers. As always, content owners can use robots.txt to signal their preferences. This is not about scraping news sites, and these updates don't give us new rights that supersede those preferences.”

The update landed inside a running argument

The update arrived during a running argument about how AI systems gather the web. Publishers have spent the past few years weighing licensing deals against crawler blocks, and robots.txt has become a live bargaining tool rather than a formality. Meta sits on both sides of that line. It runs platforms whose content others want to index, and it is one of the companies being sued over how AI training data is gathered. Both Nieman Lab and TechCrunch placed the terms episode inside that larger dispute over crawling and content licensing.

That context is why a wording change drew a response usually reserved for a product decision. The clause did not need to be used to cause a reaction. It only needed to be readable as a grant of rights, and to sit in a document that business users accept as a condition of participating.

A short sequence with a long tail

The order of events matters. The clause was added, criticised and removed inside roughly 48 hours, all before its 30 October effective date. It never took effect in the form critics attacked. That is not the same as a company reversing course after months of use, and the distinction is worth keeping straight.

What the episode does show is how quickly a terms update can be rewritten once enough people read it. Platform disputes increasingly run through documents that set the rules for access, licensing and training long before any court weighs the question. Meta's litigation record makes the stakes plain. The company won a 2025 ruling in a copyright case over training on books, but publisher suits continue, including a class action filed in May 2026 by five book publishers that alleges Meta pirated millions of books for its Llama models, per Nieman Lab. Against that backdrop, a paragraph about crawling and training is not a technical footnote. It is the part publishers watch most closely.

The same dynamic runs in the other direction. Meta argues the language simply gathered existing terms into one place, which is a common housekeeping move. Critics argue that consolidation is exactly how a new right can appear without an announcement. Both readings depend on wording that most users will never open, which is why a single sentence can set off a fight that a product launch rarely does.

The edit itself was quiet. There was no blog post to mark the revision when the training sentence disappeared from the terms. Confirmation came from reporters checking the page and putting questions to the company, which is also how the original clause drew attention to begin with. The change was visible mostly to people who read terms of service for a living.

Caveats worth keeping in view

Two readings need to be separated. The interpretation that Meta could train its AI on a publisher's site is the assessment of outside legal experts, not a statement of Meta's intent, and Meta disputes it. The company's position is that the terms described crawling for its ads and business products and changed nothing about its crawlers.

The timeline is the other point. The clause was removed two days after it was reported and 13 days before it would have taken effect. Reports should describe the sequence as added, criticised, removed, because a removed clause that never applied is a different thing from an active one.

There is also a limit on what the “test credentials” line supports. Experts said it could allow sign-ups that bypass a registration wall. Meta did not answer that question. Stating that Meta planned to breach paywalls goes beyond what any source shows.

“We've heard you and want to clarify: the crawling terms we consolidated only applied to our ads and business products, and we've clarified that in the terms themselves.” — Meta, in a statement to TechCrunch, 17 October 2026

What to watch

Three things will show whether this stays a two-day story. The first is the commercial terms page itself. If further edits appear before or after 30 October, that will say more about the company's intentions than the statement did. The second is the publisher litigation already under way, where the question of what a platform may train on is being argued with evidence rather than a clause. The third is robots.txt, the signal Meta says it still respects. How much weight that file carries in practice is the unresolved part of the whole exchange.

For newsrooms, the practical lesson is narrow and repeatable. Business accounts on major platforms now sit under terms that can change with little notice, and the changes can touch content a publisher hosts elsewhere. Reading the documents is unglamorous work. This week it was the difference between catching a clause and missing one. Meta says the removal settles the question. The litigation suggests the larger question is still open, and the next version of the terms will be the place to check.

Cover image: Wikimedia Commons "Facebook Offices Menlo Park California (34183281176).jpg" by Anthony Quintano, CC BY 2.0.