Meta is replacing the contract that covers anyone who uses its products for business, from running ads to managing a Page. The new version takes effect on October 30. It says the business authorizes Meta to crawl the websites associated with its account, to create test credentials so it can get in, to keep pulling updates, and to use what it collects to train Meta's products, which include its AI. A business that disagrees is told to stop using Meta for business.
A signature settles what a court has not
AI companies have mostly relied on the argument that reading the open web to train a model is fair use, and courts are still deciding it. News organizations are fighting OpenAI over exactly that. A contract skips the fight. If a business has agreed to let Meta train on its site, there is nothing left to argue about it, whichever way the courts go. Robots.txt is a request a website makes. A contract term is something its owner signed.
The access has a price on the open market. In March, Meta agreed to pay News Corp up to $50 million a year for three years of access to the Wall Street Journal and its other titles, for chatbot answers and for training. Every publisher with a Facebook page is now being asked for a similar kind of access at no charge, and it cannot refuse without giving up Facebook. That also weakens the next publisher's position in a licensing talk, because Meta has a free route that its business customers will have signed. Meta has made this kind of trade before: in September it cut prices on its AI coding tools for customers who share their data. Here, the price of staying on Facebook for business is the content of the website.
What is added, and what Meta already had
Meta already ran a crawler for training and improving its AI, and its 2024 ad terms let it read the links a business put in its ads, but only for those ads and with no mention of training. The new paragraph covers any site associated with the account, covers all business use of Meta, names training, and allows test credentials. Those credentials would let Meta sign up on a site the way a visitor does, and so reach pages behind a free registration. Meta has not said whether that includes paid content.
Meta's own answer is that it is gathering crawling language that was spread across several documents into one, that its crawling of publishers will not change, and that robots.txt still works. The new text does not mention robots.txt, so lawyers who reviewed it say Meta could argue the contract overrides it. Nobody knows yet which would win, and whether a court would enforce the clause at all is untested.
What it looks like for a small business
A furniture shop has a Facebook page and a few ads. Over the years it has written product descriptions, care guides and delivery rules on its website. From October 30, Meta may use all of it to train its models, and Meta's AI can then answer a customer's question about sofas from text the shop wrote, without sending the customer to the shop. The shop gets nothing for the text and has no say in where it appears.
The same applies to a school's admissions pages, a hospital group's service guides or a warehouse operator's rate sheets. Anything public on a linked site is in scope, and anything behind a simple sign-up may be as well.
What to do, and what comes next
The practical step is a few lines in robots.txt that disallow Meta's training crawler, which is named meta-externalagent. Leave the link-preview crawler, facebookexternalhit, alone or links shared on Facebook will stop showing their preview. Meta's own documentation says a third crawler, meta-externalfetcher, which fetches single links for its AI, may ignore robots.txt, so a block is not a guarantee.
If the clause holds, expect copies. Any platform that sells business tools, such as ad networks, marketplaces and email services, can add the same paragraph and collect consent from its customers instead of fighting for it. It took Google's publishers a long complaint and a regulator to get an opt-out, and Meta's business customers face the same wait. Until then, the business account is where consent to AI training is collected, and the notice about updated terms that most owners click past is the one to read.