• 7 min read
Meta removed language allowing AI training on business-linked sites, but its revised terms still do not explicitly define crawler limits.
Image: The Verge
Meta has removed the clause in its forthcoming commercial terms that explicitly allowed it to use content from business-linked websites to “provide, develop, improve, research, train, and maintain” its products. The rollback follows criticism that the language could turn a business relationship with Meta into permission for broad website crawling and AI-model training.
Meta says it consolidated terms that applied to advertising and business products and clarified them. It also says content owners can use robots.txt to signal their preferences, and that the rewrite does not create rights that override those signals. Meta says the change does not alter its crawler operation.
“We’ve heard you and want to clarify: the crawling terms we consolidated only applied to our ads and business products, and we’ve clarified that in the terms themselves. It does not change how Meta uses web crawlers. As always, content owners can use robots.txt to signal their preferences. This is not about scraping news sites, and these updates don’t give us new rights that supersede those preferences.”
The draft explicitly authorized training; the revised wording does not. The available wording does not fully describe what Meta’s crawlers may collect from sites associated with commercial users, what it may retain, or which downstream products may use collected material.
What the original clause authorized #
The earlier provision covered domains, pages, URLs, and available content associated with a company’s products or services, or made available through Meta products. It authorized Meta or an entity acting on its behalf to use automated software, including web scrapers, bots, robots, spiders, crawlers, and user agents. It listed access, indexing, caching, analysis, collection, retrieval, crawling, and periodic updates as permitted operations.
It also authorized the creation of “test credentials.” Cornell Law School professor James Grimmelmann said that wording could allow Meta to register for sites to obtain access to gated material. That is an interpretation of the draft language, not evidence that Meta used credentials to access paywalled content.
Recommended reading
LG’s TV privacy defense leaves its data lifecycle unanswered
Sergey Kuznetsov • • 10 min read
After the access authorization, the clause said Meta could use “any content” from those web locations to provide, develop, improve, research, train, and maintain the integrity of Meta products, services, and features under the commercial terms and applicable supplemental terms. The word train has been removed from the language described in the revised terms.
| Issue | Earlier commercial-terms clause | Revised terms as described |
|---|---|---|
| Covered material | Content from domains, pages, and URLs tied to a commercial user’s products or services | Terms emphasize advertising and business products |
| Automated collection | Explicitly listed crawlers, bots, scrapers, user agents, and other automated software | Meta says there is no change in how it uses web crawlers |
| Listed use of collected content | Included providing, developing, improving, researching, and training Meta products | “Train” is no longer explicitly listed |
The original rollout was set for October 30, 2026. The supplied reporting does not establish whether that effective date changed with the text, so businesses should not infer a new date from the public clarification.
Why commercial accounts made the clause unusually broad #
The draft did not read like a conventional seller-feed permission. It defined commercial users broadly enough to include accounts that advertise, market, sell, list products, build apps, create or use catalogs, manage a Page or business Group, or use a business product such as Meta Business Suite. A company can use Meta for marketing or social-account administration while maintaining a website far larger than its Meta storefront.
Meta’s stated purpose has been narrower. Andy Stone, Meta’s vice president of communications, said the company needed to pull accurate third-party product information for on-platform advertising and content recommendations, rather than scrape news sites. Meta also says site owners can use robots.txt flags to state their preferences.
The old contract text, however, combined product-information collection with access, caching, analysis, research, and training across associated domains. The wording drew objections from organizations that accept ordinary product-feed crawling as part of an advertising system. Authorization to retrieve a price or availability update is not textually identical to authorization to use all available site content to train a general-purpose model.
Matt Topic, a media and intellectual-property attorney at Loevy & Loevy, called the draft a major contractual provision to place in commercial terms.
“This is a remarkable provision to bury in [the] terms of service, given that this would give Meta access to material that would likely cost many millions of dollars in a licensing market.”
Topic also said it was unclear whether a court would enforce the term, while advising publishers to assume it could be enforceable. The reporting does not establish a definitive legal answer on either the prior clause or its replacement.
The remaining issue is collection, not the deleted word #
The revised language appears to narrow the stated purposes for publisher content, but removing “train” from a list differs from expressly excluding AI training or setting a clear collection boundary. Sam Quigley, a spokesperson for the News/Media Alliance, said the revision appeared to limit permitted uses without addressing Meta’s ability to crawl and obtain the content in the first place.
“They may have taken out training from the listed uses, but with it not being expressly ruled out, it’s still potentially on the table.”
Meta says robots.txt can signal content-owner preferences and that its crawler behavior has not changed. Yet the revised terms, as described in the reports, do not state crawler commitments as directly as the deleted clause stated collection powers. A site operator looking for a contract term saying Meta will not crawl material disallowed by robots.txt does not get that assurance from the quoted revision.
Meta has not said it will ignore robots.txt; it has said the opposite about owners' ability to signal preferences. Nor does the reporting show a new crawler deployment or a change in Meta’s collection behavior. The public explanation states a policy position, while the available description of the rewritten commercial terms does not fully specify the boundaries that prompted the backlash.
Meta’s business-agent strategy raises the stakes #
The dispute comes as Meta assembles business-facing AI products under its Enterprise Platform. In September, Meta put Muse, business agents, APIs, and coding tools under a unit led by former MongoDB CEO CJ Desai; we previously covered the enterprise stack’s still-unpriced rollout. Meta presented Muse as an agent that could serve small businesses, with free access alongside $20-per-month Power and $100-per-month Max tiers. That product direction makes website access more sensitive than it would be for a narrowly scoped ad crawler. A business agent can benefit from current pages, catalogs, support material, and operational information. The draft terms were broad enough that critics feared they could provide both an indexing mechanism for such systems and a contractual route to train them on the same material.
No supplied source establishes that Muse used the contested commercial clause, crawled the websites in question, or trained on any particular publisher’s material. The connection is strategic rather than evidentiary: Meta is promoting AI systems for commercial work while defining the website-access rights attached to its business accounts.
The timing also heightened publisher concerns. Meta has introduced Meta One, a paid subscription that allows some Facebook and Instagram users to bypass external-link limits. Some publishers have received exemptions, but the reporting says Meta did not answer questions about Meta One or access to paywalled material. Under the original draft, publishers could be charged for distribution while Meta claimed the ability to crawl their linked sites for free.
Four reports, one rollback, different tests for whether it is enough #
The four accounts agree on the central event: Meta deleted the clause that expressly named AI training as an allowed use of content from covered websites. They differ less on the facts than on the test for a satisfactory fix. Meta’s statement says the consolidated language was limited to ads and business products and did not alter crawler practice. The skeptical accounts focus on what the replacement does not explicitly say: whether collection itself is bounded and whether disallowed crawling is contractually ruled out.
These are separate layers. A term can authorize crawling, limit some downstream uses, and still leave retention, indexing, analysis, agent access, and future product uses unclear. Deleting “train” removes the clearest textual basis for claiming that a commercial user granted training permission.
Meta has withdrawn the provision most likely to be treated as a bulk AI-training license, but it has not put its robots.txt assurance into the commercial-terms language critics sought. The question for businesses and publishers is whether Meta will publish a clear, durable rule separating product-data crawling from collection rights over an entire associated website.
Frequently asked questions #
Did Meta remove AI training from its commercial terms?+ #
Yes. Meta removed language that explicitly allowed it to use content from covered domains, pages, and URLs to train its products.
Will Meta stop crawling business websites?+ #
Meta says the terms change does not alter how it uses web crawlers and that content owners can use robots.txt to signal their preferences. The supplied reporting does not establish a change in crawler behavior.
When were the original terms due to take effect?+ #
The earlier commercial-terms update was scheduled to take effect on October 30, 2026. The reporting does not confirm whether the revised text changed that date.
[Sergey Kuznetsov](https://forgeeks.net/authors/sergey-kuznetsov/)
Editor-in-Chief
Sergey Kuznetsov is Head of Product at iXBT.com, one of the largest Russian-language technology media outlets, and the founder of itzine.ru. He has spent over a decade building and running tech newsrooms. At for(geeks) he sets editorial standards and reviews what ships.