Meta drops “train” from its business-crawling terms

A narrower purpose, with crawling permission retained

The 16 October archive contains a separate sentence permitting use of collected website content to develop, research and train Meta products. In the revised preview checked on 19 October, that sentence has gone. Its replacement puts the purposes inside the crawling paragraph and specifies advertising and business products, plus integrity and security.

The revision leaves substantial access language in place: automated crawlers and test credentials, associated domains and pages, indexing and caching, and periodic updates. Comparing the texts establishes a change in the proposed contract. It does not measure what Meta’s bots have fetched.

Publishers challenged the earlier wording

The News/Media Alliance, a news-publisher trade group, objected on 16 October. Its president and CEO Danielle Coffey argued that access to Meta’s audience was being tied to broad website-scraping permissions. She raised examples beyond newspapers, including music and stock photography.

That is an interested industry’s objection to the earlier draft, not a court ruling or a review of the revised text. Its underlying question remains useful: how far does a business-platform relationship reach into material hosted on a separate website?

Meta says this clarifies existing practice

Meta spokesperson Andy Stone initially said on 16 October that crawler use was unchanged and that the update concerned accurate product information for on-platform advertising and recommendations. In his 17 October clarification below, he said the terms had been clarified to cover ads and business products. He also said the update granted no new right to override owners’ robots.txt preferences.

That matters because a robots.txt file communicates a site owner’s crawling preferences, while terms set out the contractual permission being requested. Meta’s explanation addresses how it says the two interact; a text revision alone cannot prove compliance across websites.

The crawler name still matters

Meta’s crawler documentation distinguishes several jobs. Meta-ExternalAgent includes foundation-model training; Meta-ExternalAds supports advertising and business services; Meta-WebIndexer improves Meta AI search results. Removing a word from this commercial-terms clause does not establish that the company has stopped training on web content.

The documentation tells owners to use rules for the relevant crawler in robots.txt and allows up to 24 hours for cached rules to refresh. It also names exceptions: user-requested Meta-ExternalFetcher may bypass those rules, and FacebookExternalHit may do so for security or integrity checks. A request from one agent therefore should not be reported as evidence of another agent’s purpose.

For a publisher or brand, the unresolved issue is how the narrowed permission operates alongside those separate technical controls. The useful next evidence would be verified access logs and an explicit explanation of any conflicting instructions. This comparison concerns the English (US) text; it does not establish that every language version has received the same revision.