MENLO PARK, Oct. 17 — Meta has removed the word “train” from the revised Commercial Terms taking effect Oct. 30, 2026, Crypto Briefing reported today, two days after Nieman Lab published a report in which three legal experts warned the original language let Meta scrape any website “associated with” a business account to “provide, develop, improve, research, train and maintain the integrity of” its company products. The crawling permission itself stays.

The retreat is narrower than it looks. Per Crypto Briefing, the revised version limits scraping to advertising and business product development and affirms that Meta respects robots.txt. It doesn’t say robots.txt contractually overrides the terms, and PPC Land notes the published preview itself contains no robots.txt commitment.

The Nieman Lab report, citing Cornell Law School’s James Grimmelmann among others, made the original clause legally embarrassing: it functionally handed Meta an unlicensed training corpus of every advertiser’s blog, product page, and press archive. The News/Media Alliance called the initial terms “outrageous.” Deleting one verb defuses the publisher lawsuit vector without touching the ad-stack plumbing underneath.

For a founder running a Facebook Page or an Instagram ad account, the decision window is 13 days. Acceptance is passive. Continued use after Oct. 30 counts as agreement, and there’s no partial opt-out. The levers that remain live in robots.txt, where Meta documents two distinct crawlers: meta-externalagent for AI training and meta-externalads for advertising use, per Relevant Audience.

Blocking them isn’t free. Meta’s commerce stack is widening fast, with in-ad product browsing launching Oct. 28 and the Muse agent rolled out to small businesses on Sept. 29. A site that disallows the crawlers may vanish from the surfaces where Meta is routing buying intent.

The clause survived. Only the sentence that invited a lawsuit didn’t.

Sources