Analysis
Meta quietly rewrote its commercial terms to let the company use content from any business account's website — domains, pages, and URLs — to "research, train, and maintain" its AI products, then removed the language within weeks after legal experts flagged how far it reached, Meta confirmed this week. A Meta spokesperson said the change was procedural: "the crawling terms we consolidated only applied to our ads and business products" and "it does not change how Meta uses web crawlers." The company's fuller statement on the removal was blunter: "We've heard you."
The provision surfaced in early October, when Meta alerted business-account holders to updated terms through a blog post and the Meta Business Suite. Nieman Lab reported that legal experts reviewing the language said it could let Meta scrape the websites of any publisher or business running a Meta ad or Business Suite account and feed that content into its AI training pipeline — a provision one expert called "remarkable" for how broadly it defined usable content. Pulse covered the original clause when it surfaced, noting the scope of what it would have allowed.
What Meta Actually Walked Back
The removed language specifically authorized Meta to pull content from a business account's "domains, web pages, or URLs" for use across its products — not just ads delivery, but model training itself. That distinction matters because Meta has spent the past year defending its AI-training practices in court: the company defeated a high-profile copyright lawsuit over training on books last year, and separate cases against Meta and other AI labs over scraped training data are still working through the courts. A clause granting Meta an explicit contractual right to scrape business customers' own sites would have been a much cleaner legal foothold than the ambiguous "publicly available" standard most AI labs currently lean on.
The timing also lands next to Meta's own AI ambitions. Pulse has reported that Alexandr Wang is steering a $100B-plus AI bet at Meta built around its new Muse Wonder frontier model — a model that needs training data at scale, and one more reason critics read the now-deleted clause as more than boilerplate.
For founders and marketers running ad accounts or Business Suite pages, the practical risk is narrower than headlines suggested but not zero: Meta's own statement that its crawler behavior "never changed" implies the AI-training use case critics worried about may already be covered under existing, broader terms — pulling six words from a contract doesn't retroactively change six years of crawling behavior. Critics who flagged the clause are not declaring victory; they're treating the removal as a reason to keep reading the next terms update closely, not the last one.
Meta has not said whether a narrower version of the clause will reappear in a future terms update, or whether the removal affects any content already ingested under the deleted language.