Meta has revised its commercial terms after language suggesting the company could use content from websites linked to business accounts for AI development triggered concerns among publishers and legal experts.
TechCrunch reported that Meta’s earlier wording said the company could use content from associated domains and URLs to provide, develop, improve, research, train its products and services. The breadth of that provision raised questions about whether publishers operating Meta business accounts could inadvertently give the company permission to scrape their websites for AI training.
Meta has now removed the most controversial language, but the episode highlights a growing fight over how AI companies obtain access to professionally produced online content.
Meta removes explicit AI training language
The revised commercial terms no longer contain the provision linking website content directly to model training.
The Verge reported that Meta removed language that suggested it could scrape websites, including news publishers, after the wording attracted scrutiny. The change specifically removes “train” from the list of permitted uses and narrows the language toward Meta’s advertising and business products.
Meta said the earlier terms did not represent a new approach to web crawling. In a statement reported by TechCrunch, the company said the consolidated crawling terms only applied to our ads and business products and maintained that the revision does not change how Meta uses web crawlers.
The company’s updated commercial terms are scheduled to take effect on October 30, 2026.
Publishers feared the terms could weaken scraping controls
The concern was not simply that Meta operates web crawlers. Search engines, AI companies and other online services routinely use automated systems to retrieve public web content.
The dispute centered on whether accepting Meta’s commercial terms could create contractual permission broader than the technical access publishers normally control.
The Verge reported that legal experts had questioned whether the earlier wording could potentially override restrictions publishers place through robots.txt files. Meta, however, has maintained that publishers’ crawler restrictions remain respected.
That distinction matters because robots.txt has become an increasingly important boundary in the AI training debate. Publishers can use it to tell automated crawlers not to access certain material, even though its legal enforceability and implementation vary.
AI companies and publishers are fighting over access to the web
Meta’s revision arrives as publishers are becoming more aggressive about controlling how their reporting is used by generative AI companies.
The dispute is increasingly moving beyond traditional copyright litigation and into contracts, crawler permissions and licensing arrangements. If a platform can obtain broad content rights through business terms, publishers could lose some negotiating leverage over separate AI licensing deals.
Meta’s retreat from explicit training language does not eliminate every question about its ability to crawl websites associated with business accounts. The revised language still provides Meta with permissions connected to its advertising and business products.
For founders and AI product teams, the episode shows why training-data strategy is becoming partly a product-policy and contract-design problem, not simply a technical one.
The web may look publicly accessible to a crawler, but access, permission and reuse are becoming three increasingly different concepts.
For publishers, those distinctions determine whether their content becomes licensed data, blocked material—or another input absorbed into the infrastructure being built on top of the open web.
