Meta Removes Wording That Seemed to Permit Scraping News Sites for AI

Meta has removed wording from its Commercial Terms that appeared to let it collect content from outside websites, including news sites, to train its AI. The Verge reported the change on Oct. 17.

The removed clause said Meta could use any content from domains, web pages, or URLs to provide, develop, improve, research, train, and maintain Meta products. That text no longer appears on Meta's commercial terms page.

Meta spokesperson Daniel Roberts said the combined crawling language was only meant to cover ads and business products. He said the update does not change how Meta's web crawlers operate, and does not give Meta new rights that override a site owner's robots.txt preferences. Crawlers are automated systems that fetch pages from the open web. Robots.txt is the standard file sites use to signal what crawlers should avoid. In practice, crawler behavior depends on crawler code, robots.txt signals, and any separate contract permission.

Roberts had earlier disputed reporting on the business-terms update. The Verge described that dispute on Oct. 16, after Nieman Lab laid out the draft language.

The earlier draft, set to take effect Oct. 30, said continued use of Facebook, Instagram, or WhatsApp would count as consent by a business account for Meta to scrape its website. Scrape here means automatically collect pages at scale. Legal experts said that wording would permit scraping of news sites, according to parallel coverage from Nieman Lab and Editor and Publisher. For news outlets that run both a website and a business account on Meta platforms, that link was the issue they raised.

Meta says the Commercial Terms updates take effect Oct. 30, 2026. Meta says the terms cover access or use of Meta Products for a business or commercial purpose, including using ads, selling products, developing apps, managing a Page, managing a Group for business purposes, or using measurement services. The current Licenses section covers content shared, posted, or uploaded on or in connection with Meta Products. For U.S. residents or businesses, most disputes must go to binding arbitration on an individual basis, with no class actions and no jury trials.

The removed language mixed two different permission models. One is the platform license, which covers content a business uploads directly to Meta services. The other is web-wide collection, in which a crawler visits third-party sites under technical controls and copyright law. Mixing them left engineering and legal teams responsible for data governance, model provenance, and publisher relations without a clear line to audit.

The broader context here helps explain why outside lawyers reacted quickly. Terms of service have become another place where AI training data is contested, alongside robots.txt, opt-out headers, and licensing deals. Wording can be read broadly even when a company says it intended no operational change, which is why the consent-by-continued-use phrasing drew attention.

In my view, the reversal lowers ambiguity without settling the larger question. Publishers still need clear, machine-readable ways to state crawling and training preferences. Model builders still need documented permissions that can be checked and withdrawn. A short clause cannot resolve that, but precise terms tied to observable crawler behavior can help.

As a practical matter, treat platform terms like versioned software. Note effective dates, and review robots.txt, crawler logs, and contract records together. My kids grew up assuming anything posted online would be copied widely, and for simple distribution they were largely right. Training data is where that old assumption is now being worked out line by line, and clearer rules should make reuse more reliable for everyone.