Meta removes ToS language that suggested it could scrape websites "associated with" business accounts, saying it applied only to "ads and business products" (Terrence O'Brien/The Verge)

Meta removed terms-of-service language referring to websites “associated with” business accounts. The wording mattered because it appeared broader than a rule covering Meta’s own advertising and business tools. It suggested that websites connected to a business account might fall within the company’s claimed permissions.

For example, a news organization with a business account could potentially have been viewed as linked to that account. Experts said the wording could therefore have supported scraping the organization’s public website. Scraping means collecting website information automatically, often across many pages.

Meta now says the relevant language applied only to “ads and business products.” That clarification narrows the apparent scope. The change does not settle every question about AI training, copyright, or consent, but it removes wording experts viewed as potentially enabling broad collection from news sites.

Meta’s ads and business products are commercial tools offered to organizations, advertisers, and other businesses. They can include systems for creating ads, managing campaigns, and handling business activity on Meta’s services. The provided article does not list specific products, so this description uses established industry knowledge.

Business accounts act as the control point for those tools. An organization can use an account to create advertising campaigns, manage pages or other business assets, and work with Meta’s commercial features. The account identifies the business relationship, while the products provide the functions.

This distinction is central to Meta’s explanation. Meta said the removed wording applied only to “ads and business products,” rather than to every website associated with an account. That narrower reading would focus on Meta’s commercial tools, not automatically on an outside news website connected to the business.

The phrase “websites associated with” business accounts was broad because “associated with” can describe an outside relationship. It could potentially include a company’s own website, a publisher’s site, or another webpage connected to a business account. By contrast, “ads and business products” points to a narrower group of Meta-controlled commercial services.

For example, a news organization might operate a business account while publishing articles on its own domain. A broad reading could treat that domain as associated with the account. A narrow reading would cover the organization’s Meta advertising or business tools, but not necessarily its website content.

The article does not provide a list of covered sites or a measured scale. It reports that experts believed the wording allowed scraping of news sites to train AI. Meta removed the language and said it applied only to ads and business products, reducing the apparent reach of the terms.

Web scraping is the automated collection of information from websites. A program, sometimes called a crawler or scraper, requests webpages and extracts readable elements such as text, links, images, or metadata. It can repeat this process across many pages much faster than a person could.

For example, software could visit a news site’s article pages, copy their headlines and text, and store the results in a database. A company might then organize the material, analyze it, or use it as training data. The exact access method and legality depend on the site, permissions, and applicable law.

In this case, experts said Meta’s removed wording appeared to allow scraping of news sites for AI training. Meta said the language applied only to “ads and business products.” That statement narrows what the company says its terms covered, but it does not explain every technical or legal limit on collection.

News websites can be valuable for AI training because they contain large collections of human-written language. Articles provide examples of vocabulary, grammar, facts, topics, and different writing styles. Public webpages can also offer broad and changing information. These points are established context for AI training, although the short article does not detail Meta’s training datasets.

For example, a scraper could collect thousands of publicly accessible news articles and place their text into a dataset. An AI system could use that material to learn patterns in language. The system would not need a person to copy each article manually; automated software could gather pages at scale.

The article reports that experts said Meta’s terms appeared to allow scraping news sites to train its AI. Meta removed the relevant language and said it applied only to “ads and business products.” The dispute shows why publishers may closely examine wording that could cover their webpages.

If broad terms were interpreted as permission to scrape content, publishers and website owners could see their webpages collected for uses they did not expect. Their articles, images, or other material might enter large datasets without a separate agreement. The core concern is control: unclear wording could turn a business-account relationship into apparent permission over an outside site.

For example, a news publisher could maintain a Meta business account while hosting stories on its own domain. Under a broad reading of “websites associated with,” software might collect those stories for AI training. The publisher might then need to challenge the interpretation or seek clarity about how its content was used.

The article does not describe actual damages, payments, lawsuits, or specific publisher responses. It reports that experts saw the language as allowing scraping of news sites. Meta removed it and said it applied only to ads and business products, potentially reducing uncertainty while leaving broader AI-content disputes unresolved.

Terms of service are rules governing use of a platform and may describe what the platform or users permit. Copyright law can separately protect original writing, images, and other creative work. Consent is permission from the rights holder or another authorized source. These systems can overlap, but one broad clause does not automatically answer every copyright or consent question.

For example, a publisher might accept Meta’s terms to operate a business account while retaining rights in articles on its own website. Whether Meta could collect those articles for AI training could depend on the wording, the relationship between the services, copyright exceptions, contracts, and local law. Public access alone is not necessarily permission for every reuse.

The article reports that experts viewed Meta’s wording as allowing news-site scraping. Meta removed it and said it applied only to “ads and business products.” That change narrows the contractual language, but the excerpt does not resolve copyright law, consent requirements, or future disputes.