Meta has removed the word “train” from its updated commercial terms, deleting language that appeared to let the company crawl the websites of businesses that use its products — including news publishers — and use what it found to train AI models. Why it matters: after days of backlash from publishers and legal experts, the broadest claimed permission is gone, but Meta’s crawling rights for its ads and business products remain, and the revised terms still take effect on October 30.
October 17, 2026
What happened
Nieman Lab reported on October 15 that a new paragraph in Section 1 of Meta’s Commercial Terms authorized the company to use automated software — scrapers, bots, crawlers and similar tools — and created test credentials to access, index, cache, collect and crawl the domains associated with a business’s products or services, or made available through activity on Meta’s products. A following sentence said Meta could use any content from those domains to provide, develop, improve, research, train and maintain its products. The terms are scheduled to take effect on October 30, 2026, replacing the version dated March 6, 2024.
On October 17, The Verge reported that Meta had removed the sentence containing the training language. The crawling permission that remains covers Meta’s advertising and business products and the security of its services. A Meta spokesperson said the company had clarified in the terms themselves that the consolidated crawling language applies only to its ads and business products, that the change does not alter how Meta uses web crawlers, and that content owners can still use robots.txt to signal their preferences.
Why publishers pushed back
Meta’s definition of a business customer is broad: it covers anyone using ads or marketing messages, selling or listing products, developing apps, using a product catalog, or managing a Page or a business Group through tools such as Ads Manager or Meta Business Suite. As Nieman Lab noted, that includes almost every news publisher. Legal experts who read the original text told Nieman Lab it could support scraping news sites for AI training, and acceptance was passive — continued use of Meta’s business products counted as agreement, with no opt-out short of leaving. The News/Media Alliance condemned the original terms, and Meta communications executive Andy Stone responded on October 16 that the update was meant to ensure Meta could pull accurate information about products for sale on third-party websites for on-platform ad and content recommendations, not to scrape news sites.
What did not change
The October 30 effective date stands, and the terms still authorize Meta to crawl domains tied to a business’s products or services using automated tools and test credentials, and to pull updates periodically — now stated for advertising, business-product and security purposes. The terms themselves still do not mention robots.txt; Meta’s assurance that robots.txt preferences are respected comes from its statements, not from the contract text. Publishers who were worried the contract could be argued to outrank a robots.txt block have less to point to than they did on Thursday, but no new contractual guarantee either.
Why it matters
For anyone running a website — including this one — the episode is a clean test of how AI training data gets collected in 2026. Other AI companies are signing paid licensing deals with publishers or fighting them in court. Meta’s first draft tried to source a similar permission from the terms businesses already accept to reach Facebook and Instagram audiences, and it took about two days of public pressure to get the training sentence deleted. The practical takeaway for site owners is unchanged: keep your robots.txt deliberate and current, because that file — not a platform’s press statements — is the control you actually hold.
Frequently asked questions
Did Meta admit it was scraping news sites to train AI?
No. Meta said the clause concerned product information for its ads and business products, that it was a consolidation of existing terms, and that its crawling practices have not changed.
What exactly was removed?
The sentence that allowed Meta to use content crawled from business-associated domains to “train” its products. The crawling authorization itself remains, with narrower stated purposes.
When do the revised terms take effect?
October 30, 2026. The removal of the training language happened before that date, while the terms were still in their pre-launch form.
Does robots.txt stop Meta’s crawlers?
Meta says content owners can use robots.txt to signal their preferences and that the updated terms give it no new rights that supersede those preferences. The terms text itself does not mention robots.txt.
Sources: The Verge, Nieman Lab, The Register, PPC Land, Android Headlines