Meta has removed a sentence explicitly permitting content from websites associated with business users to be used to “train” its products from its pending Commercial Terms. Archived copies dated October 16 and October 17, 2026, confirm the change, which leaves crawling authorization in place but revises its stated purposes. Both versions carry an October 30, 2026, effective date. (web.archive.org)
The Verge reported the removal on October 17. Meta spokesperson Daniel Roberts told the outlet that the revision clarified the terms’ application to advertising and business products, rather than changing how Meta uses web crawlers. Legal experts interviewed separately by Nieman Lab said the revised wording remained broad. (theverge.com)
What changed in the wording
The October 16 preview contained two distinct elements: authorization to crawl associated websites and periodically retrieve updates, followed by a separate sentence permitting use of their content for a range of purposes. That second sentence expressly included training among uses connected to Meta Company Products, services and features. (web.archive.org)
In the October 17 capture, that separate sentence is gone. Instead, the crawling provision runs into language describing purposes involving “advertising and business products,” alongside maintaining the integrity and security of Meta Company Products. The revised passage no longer expressly lists training. (web.archive.org)
The comparison is therefore more than a single-word deletion. It changes how the provision describes the purposes of collecting website content, while preserving authorization to crawl associated domains, pages and URLs and return for updates. It is not a removal of the website-crawling provision as a whole. (web.archive.org)
When the revision happened
Nieman Lab places the revision on Friday afternoon, October 16. The inspected archive captures establish that the disputed sentence appeared in the October 16 version and was absent from the October 17 version. They document a before-and-after sequence, not the exact moment Meta edited its page. (niemanlab.org)
That distinction also separates the publication of the revised preview from the terms taking effect. October 17 is the date of the follow-up reporting and the later archive capture; October 30 remains the effective date printed in both inspected versions. The evidence does not show that the revision accelerated that date. (theverge.com)
The original Nieman Lab report, published October 15, examined concerns that the pending terms could authorize AI training using content from websites associated with business accounts. It supplies background to the dispute, rather than evidence of the subsequent removal. Readers following that earlier account now have a materially different text to consider. (niemanlab.org)
Meta’s explanation and the remaining questions
Roberts said website owners could continue signaling crawling preferences through robots.txt and denied that the update gave Meta new rights overriding those preferences. Those statements describe the company’s position; they are not an independent audit of crawler behavior or a legal determination about the revised contract. (theverge.com)
Matt Topic, a media and intellectual property attorney at Loevy & Loevy, told Nieman Lab that permission to develop business products could still encompass training. Cornell Law School professor James Grimmelmann likewise questioned the breadth of the remaining language and noted that the wording identifying associated websites had not been restricted. These are attributed legal assessments, not court findings. (niemanlab.org)
Nieman Lab also reported that Meta had not answered its specific questions about the revision when its follow-up was published. That outstanding request should not be confused with an absence of any company response: Meta had supplied a statement explaining its position. (niemanlab.org)
A wider debate over content access
The dispute comes as publishers and infrastructure providers seek clearer distinctions between different uses of online content. On September 15, Cloudflare announced separate controls for search, AI training and AI agents. It described its Disallow AI Training setting as a way for websites to refuse training use while remaining discoverable in search. Those are Cloudflare’s stated capabilities, not findings about Meta’s compliance. (cloudflare.com)
Meta has also pursued publisher partnerships for a different, explicitly described purpose. In March, it announced additional international news sources for Meta AI, including News Corp, Le Figaro, Prisa and Süddeutsche Zeitung. Its announcement described real-time answers with links to publishers’ articles; it does not establish what the disputed Commercial Terms authorize for training. (about.fb.com)
Together, those developments illustrate why access, training and displaying information should not be treated as interchangeable. For this revision, the defensible conclusion is narrower: the express training sentence was removed, while crawling permission remained. Whether the remaining language encompasses particular AI uses is still disputed in the reporting. (cloudflare.com)
Verification note: The before-and-after comparison above uses directly inspected archived Meta previews. The live Commercial Terms and preview pages redirected to login during retrieval, so their current display could not be independently checked. The archived comparison confirms the revision, but does not resolve that remaining live-page verification requirement.
