What Changed
Meta updated its commercial terms of use on October 17, 2026.
The update removed a clause that allowed the company to scrape sites.
The clause had applied to business accounts and covered any content from domains, web pages, or URLs.
The removal clarifies that Meta no longer has blanket rights to use such content for AI training.
The new terms do not alter Meta’s existing crawling practices.
The Removed Clause
The old clause stated that Meta could use any content from a business account’s domains, web pages, or URLs.
It said the content could be used to provide, develop, improve, research, train, and maintain Meta products and services.
Legal experts called it a “remarkable provision” that could let Meta scrape any publisher’s site.
The clause was not limited to public data; it could include proprietary or copyrighted material.
Its removal removes the ambiguity about the scope of data Meta could collect.
Meta's Explanation
Meta said it heard concerns from users and wanted to clarify its terms.
The company stated that the crawling terms only applied to ads and business products.
The updated terms reflect that explicitly, removing the broader language.
Meta’s spokesperson said the change does not affect how Meta uses web crawlers.
The company maintains its existing crawling policies and continues its current approach.
Background on Scraping
Web crawlers are programs that automatically visit web pages.
They read the content of each page and store it for later use.
Crawlers follow links to other pages, building a large dataset.
Polite crawlers respect robots.txt files; aggressive crawlers ignore them.
The volume of data collected can be very large, depending on the crawler’s speed and scope.
Background on AI Training
AI models learn by seeing many examples.
Data is fed into a training algorithm that adjusts internal parameters.
The process repeats thousands of times to improve predictions.
Large language models use vast amounts of text from books, websites, and other sources.
The quality and diversity of training data directly affect model performance.