Meta has rewritten part of its commercial terms after the wording appeared to let it use content from business websites to train its AI models. The earlier text said material its crawlers collected from websites linked to business accounts could be used to “provide, develop, improve, research, train, and maintain the integrity of” Meta products. The new version drops “train” and limits that use to advertising and business products.
That sounds like a small edit, but for publishers it touches a bigger worry. Many news organizations, retailers and media brands run Facebook and Instagram business accounts that point to their websites. A clause letting Meta crawl those sites and use the content to train its products, with no limit on which products, could be read as broad permission to feed journalism into its AI models. Legal experts raised that reading. Meta says it was never the intent and that the terms were always about its ad and business tools, not a new way to scrape news sites.
The revision answers the most obvious complaint, but it leaves other questions open. Meta can still crawl those websites. The company says site owners can set their preferences through robots.txt, the long-standing file sites use to tell bots what to stay away from. Robots.txt is a convention, not a law, and legal experts point out that the updated terms don’t explicitly require Meta to follow it. The terms also don’t define “advertising and business products” very precisely, especially at a company that is putting generative AI into its ad tools, chatbots and business messaging.
Publishers’ caution is understandable. One analysis of 102 major US publishers found that 42% block crawlers used for AI training, compared with 25% that block crawlers used for AI search. The gap suggests publishers will tolerate bots that might send readers back to them, but not ones that take content and give nothing back. Infrastructure companies such as Cloudflare have also started blocking AI crawlers by default, which has made opting out the standard position rather than a protest.
Meta has been here before. It already trains on public Facebook and Instagram posts in some regions, and users have complained that objecting to that is harder than it should be. Meanwhile, publishers increasingly want to be paid for their work. Licensing deals with AI companies are becoming common, and Amazon has reportedly looked at a marketplace where media companies could license content directly to AI developers.
Removing one word buys Meta some goodwill. What would actually reassure publishers is a commitment to honor opt-outs and an explanation of where crawled content ends up, and Meta has offered neither.
