Meta removes AI-training sentence from new business terms after scraping backlash; rest of crawling clause under dispute

Meta has removed a sentence from its new commercial terms that legal experts said could let it scrape the websites of any business with a Meta account and use the content to train AI. The change came on 17 October, a day after the wording drew criticism from publishers and lawyers, and two weeks before the terms take effect on 30 October 2026. Meta says the language only ever covered its ads and business products and "does not change how Meta uses web crawlers." Critics say the part that remains still reaches far, so the dispute is not settled.

Key points

• Meta's revised commercial terms take effect on 30 October 2026 and apply to anyone using its business products, including Pages, ads, catalogues and Business Suite.
• The text let Meta use automated software to crawl "domains, web pages, or URLs" linked to a business customer and use the content to "train" its products; that final permission was removed on 17 October after criticism.
• Meta first said any suggestion of a change in how it crawls was "false," then said it had "heard you" and clarified that the terms apply only to ads and business products.
• The News/Media Alliance called the original terms "outrageous and probably illegal"; a law professor told The Register the wording raised procedural and substantive problems.
• A later Nieman Lab headline says legal experts still see broad scraping permissions after the edit; we could not open that article to check the reasoning.

What happened

The Register reported on 16 October that Meta had revised its commercial terms of service, with the new language taking effect on 30 October. The terms define business customers broadly: anyone "using ads or marketing messages, selling or listing products, developing apps, creating or using a product catalog, managing a Page, managing a Group for business purposes, using any business product, such as Ads Manager or Meta Business Suite."

The added text, as quoted by The Register, said the business "authorize[s] us (or any entity acting on our behalf) to use automated software (such as web scrapers, bots, robots, spiders, crawlers, and user agents) and create test credentials to access, index, cache, analyze, collect, retrieve, or crawl the domains, web pages, or URLs" associated with its products or services. A second paragraph said Meta "may use any content from such domains, web pages, or URLs to provide, develop, improve, research, train, and maintain the integrity of" its products and services.

Nieman Lab, which first raised the alarm, interviewed legal experts who called the provision remarkable and said it could let Meta scrape the site of any publisher holding a Meta business account and train AI on it, TechCrunch reported. The Register adds that almost every news publisher has at least one such account, and that at least a million businesses use Meta's AI agents in WhatsApp and Messenger.

Meta's first answer, from communications chief Andy Stone and spokesperson Daniel Roberts, was that there is no change to how Meta uses web crawlers and that any suggestion otherwise is false, according to The Verge. On 17 October The Verge reported that the "Meta may use any content..." sentence no longer appears on the commercial terms page. Roberts said: "This is not about scraping news sites, and these updates don't give us new rights that supersede those preferences", referring to robots.txt files in which sites state what crawlers may do.

We tried to read the live terms page directly, but it did not load for us without a browser; the wording above is as quoted by The Register and The Verge.

Background

Terms-of-service rewrites that grab more rights than intended are not new. Eric Goldman, a professor at Santa Clara University School of Law, told The Register that earlier flare-ups, such as GeoCities in 1999 and Instagram in 2012, did not lead to disaster, and that drafters sometimes solve one problem and accidentally claim rights that upset people. He said this looked like that pattern.

AI training on web content is also a live legal fight. TechCrunch notes Meta won a high-profile copyright lawsuit last year, while other cases against AI companies, Meta among them, continue. Robots.txt, the plain-text file where a site lists which crawlers it allows, is the usual way to signal preferences, but it is a convention, not law. Goldman told The Register that the law does not necessarily compel anyone to respect it, and that the earlier text did not expressly say Meta would.

The two readings

Critics' reading. Customers who want to keep using Meta's ad and business tools would have had to accept crawling of their own websites and use of the content for training, possibly overriding robots.txt. Danielle Coffey, president and CEO of the News/Media Alliance, said, before the edit, that the company was withholding access to its audience unless businesses allow Meta to scrape their content, and that the terms must not be allowed to take effect on 30 October. Goldman raised a contract-law point: customers who have prepaid for services may have no real option to reject the change except leaving, and then the question is whether Meta would refund them.

Meta's reading. The company says it merely consolidated crawling language already found in its ads and business terms, that it changes nothing about its web crawlers, and that robots.txt preferences still stand. The removal of the training sentence is consistent with that account, but it is also what you would expect from a company responding to a backlash; the text alone cannot tell the two apart.

What it means

For publishers and small businesses, the practical point is that agreeing to platform terms can carry rights over the content of your own website, not just your posts on the platform. Anyone with a Meta business account should read the terms before 30 October, keep their robots.txt up to date, and check whether the crawler they want to restrict is one Meta identifies and honours. For developers building on Meta's APIs, the clause about creating "test credentials" to access linked pages is worth reading closely.

For the AI industry, the episode shows how quickly contract language about training data draws scrutiny, and how a robots.txt promise in a press statement is not the same as a promise in the contract. Goldman argued that Meta has little goodwill left to draw on when readers interpret its contract language, which is his opinion, but it explains why the wording drew such a fast reaction.

What we don't know yet

• What the live terms say in full after the edit. Nieman Lab's follow-up headline says legal experts still see "broad scraping permissions" despite the changes, but we could not read it.
• Whether the first paragraph, which authorises crawling, remains word for word. The Verge reports only the training sentence as removed.
• Whether Meta will put its robots.txt commitment into the contract text itself.
• Whether any regulator or publisher group will challenge the terms before they take effect.

What to watch next

Watch for the final text on 30 October, any formal complaint from publisher groups such as the News/Media Alliance, and whether Meta publishes crawler documentation that matches its statement. We will add a dated update if the terms change again.

Sources

• Meta commercial terms (live page; wording taken from the reports below)
• The Register: Meta grants itself right to scrape content from business customers for model training
• The Verge: Meta removes language that may have allowed it to scrape news sites
• TechCrunch: Meta removes language that raised scraping and AI training concerns
• Nieman Lab: Meta's new commercial terms will allow the scraping of news sites, legal experts say