Overnight, Meta rewrote the clause that had publishers worried

Two captures, one word fewer

Friday, October 16, 6:20 p.m. Paris time. The Wayback Machine saves the page for Meta's upcoming commercial terms. It says Meta can use the content of business customers' websites to provide, develop, improve, research and "train" its products. The old clause reads: "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."

Saturday, 5:19 a.m.: a new capture. In the US English version, "train" is gone.

We compared the seven captures of that page archived since October 5, one of which we triggered ourselves on Saturday morning. The first five carry the old wording, the last two the new one. We found no announcement from Meta about the change, and the three articles that broke the story (Nieman Lab, Ars Technica, The Register) all quote the old version.

Think of a clause edited in a lease just before signing: the tenant who read the first draft has no reason to read the second.

What Meta asks of business accounts

Meta Commercial Terms apply to anyone who uses Meta's services for commercial purposes: advertising, managing a Page, selling or listing products, using Business Suite. The new version takes effect on October 30, 2026, a date stated plainly on Meta's page.

It adds a clause. The customer lets Meta use automated software (bots, crawlers) and "create test credentials" to browse domains "associated" with its products or services, or made accessible through the use of Meta's products. It also lets Meta return regularly to pick up updates. On our reading of the text, a Facebook Page that links to a website is enough to bring that site into scope.

Nieman Lab, Harvard's journalism lab, broke the story on Thursday, October 15 after a reader tip, and had three lawyers read the text. James Grimmelmann, a professor of digital law at Cornell, read it as permission to train Meta's AI on a news site's entire content. Matt Topic, an intellectual property lawyer, called it "remarkable" to bury such a provision in terms of service, given that access of this kind would, in his view, be worth millions of dollars in a licensing market.

Ars Technica pointed out that the text never mentions artificial intelligence at all. Everything hung on a single occurrence of the word "train."

The rewritten sentence

In the US English version served today, the end of the sentence is different. "Train" no longer appears, and the purpose narrows: the crawlers now serve to "provide, develop, improve, research, and maintain advertising and business products." That means, in the text's own examples, helping create ads, giving users information about a business's products, and connecting it with potential customers. They also serve to "maintain the integrity and security" of Meta's products.

The new wording lines up with Meta's defense. On Friday, spokesman Andy Stone said, as PPC Land reported his message, that the update was meant to pull reliable information about products sold on third-party sites for ad recommendations, not to scrape news sites. The contract language now matches the spokesman's argument.

One question the text leaves open: could "develop" and "improve" advertising products cover AI tools that generate ads? The new version says neither yes nor no.

What didn't move

The opening of the clause is intact, word for word. Meta can still send its crawlers to sites associated with a business account, return regularly, and create test credentials. Grimmelmann noted that such credentials could technically let Meta sign up as a fake user to get past a site's registration wall. Meta declined to tell Nieman Lab whether they'd be used to get through paywalls.

There's still no opt-out box. Continuing to use Meta's products after notice counts as acceptance, and anyone who disagrees "may stop" all commercial use. The robots.txt file, the "keep out" sign a site posts for bots, appears nowhere in the text. Meta says publishers can still use it; Eric Goldman, a law professor at Santa Clara, told The Register the contract doesn't commit to honoring it.

Stephen Beale, who runs a small steampunk site, found the notice in his Meta Business Suite dashboard. His community lives on Facebook. He told Nieman Lab he was considering closing his accounts before October 30.

Déclic is in the same position. Our articles are shared on a Facebook Page and an Instagram account, posted through a scheduling tool, so these terms apply to us directly, as they do to any outlet distributing its links there.

In Europe, the page still says "entraîner"

The document is the same for everyone. It contains its own clauses sending some disputes to Irish courts, plus a reference to the EU's Digital Markets Act (DMA). It provides no European exception for the crawler clause.

The rewrite, though, hasn't traveled. On Saturday morning, from a connection in France, the French-language page still said Meta may use this content to "entraîner" its products (French for "train"). The German page and the British English page carried the old wording too. Only the US English version had the new one, and the text doesn't say which language prevails.

It's not the first time Meta has reworked a contested feature quickly. In July, the company pulled a feature that generated images from Instagram accounts within forty-eight hours, as we reported in our piece on Muse, its children's mascot.

Links, meanwhile, are counted

The clause lands just as Meta tightens the other end of the pipe. With its Meta One subscription, launched in September, the company is testing a cap on free external links for business Pages: two a month, according to Press Gazette, with plans at $49.99, $149.99 or $499.99 to post more. News publishers would be exempt for now, according to the British outlet, though some small publishers have already hit the cap.

Nieman Lab sees a paradox: charging for links to content that crawlers could come and fetch for free. The paradox holds even without the word "train."

The new terms take effect on October 30. Until then, nothing stops Meta from editing them again. It has just done so overnight, and the date at the top of the page hasn't moved.