PolicyLaw, regulation & governance
Meta drops 'train' from new business terms clause letting it crawl advertisers' websites
Meta deleted the word 'train' from a new commercial-terms clause letting it crawl websites linked to businesses using its ad and business tools, after legal experts warned the wording could amount to AI-training consent buried in terms of service.
From the TechCrunch AI article of October 17, 2026 and cited sources: The original clause was set to take effect Oct. 30, 2026. [3] [4]
SOURCES[3] PPC Land: Meta's new terms let it crawl business websites… ppc.land[4] Nieman Lab: Meta's new commercial terms will allow the… niemanlab.org
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- Original source
- TechCrunch AI article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- October 17, 2026 article published
- Covered
- Oct 18 edition
- Evidence
- 10 sources across 10 sites
- 1TechCrunch: Meta removes language that raised scraping and…techcrunch.com
- 2The Verge: Meta removes language that may have allowed it…theverge.com
- 3PPC Land: Meta's new terms let it crawl business websites…ppc.land
- 4Nieman Lab: Meta's new commercial terms will allow the…niemanlab.org
- 5FourWeekMBA: Meta cuts AI-training clause from its Oct. 30…fourweekmba.com
- 6Meta for Developers: Meta Web Crawlersdevelopers.facebook.com
Reach: not measured yet
The original clause was set to take effect Oct. 30, 2026. It would have let Meta's crawlers, bots and similar agents access domains linked to any business using Facebook Pages, Instagram, WhatsApp, Messenger, Ads Manager or Business Suite, and use that content to develop, research and train Meta products. Businesses accept the terms simply by continuing to use these products. Lawyers told Nieman Lab this could allow Meta to scrape news sites for AI training, since nearly every publisher has a Meta business account. [3] [4]
About two days after the clause was flagged, Meta removed 'train' and tied the permitted uses to advertising and business products, plus integrity and security. Meta said the clause had only ever covered those uses and that its crawlers work the same as before. [1] [2] [5]
The edited terms still allow crawling of linked domains. Robots.txt remains the practical control. Meta's documentation separates crawlers by purpose: Meta-ExternalAgent collects data for foundation-model training and Meta-ExternalAds serves ads uses. The link-preview crawler and the fetcher for user-requested pages may skip robots.txt in some cases. The terms don't say robots.txt settings override the contract. [3] [5] [6]
the edit removes explicit training rights, but crawler authorization on linked domains remains.
Read the full assessment
Implication: publishers and site operators should treat platform terms like licensing contracts and enforce preferences per crawler in robots.txt, since the terms don't say robots.txt overrides them.
Sources are gathered; writing the dossier now.
Executive brief
Meta wrote a clause into its commercial terms, set to take effect on Oct. 30, 2026, that let it crawl the websites of any business using its ad or business tools. It could then use that content to "train" its products. Almost every news publisher has a Meta business account, so legal experts told Nieman Lab the clause amounted to consent to AI-training scraping hidden in terms of service. Rejecting it meant stopping use of Meta's business products. About two days later, after pushback, Meta removed the "train" wording and narrowed the clause to ads and business products. Meta says how its crawlers work has not changed.
What changed and event timeline
Previous terms in force
The earlier commercial terms had no crawling clause. This is the baseline for the new paragraph added to Section 1 ().
Nieman Lab flags the clause
Three lawyers said the wording could allow scraping news sites for AI training. Attorney Matt Topic called it remarkable to bury something like this in terms of service (;).
Meta pushes back
Andy Stone, Meta's communications chief, said the update was meant to pull accurate information about products for sale. Meta denied any change to how it crawls the web (;).
Provision removed
The Verge reported the edit. Meta said the clause only ever covered ads and business products and that crawler behavior is unchanged (;).
Revised terms take effect
Businesses accept the terms by continuing to use Meta's products. There is no partial opt-out ().
Capabilities and access
No model is involved. This is a contract change about crawling.
Read the full section
No model is involved. This is a contract change about crawling.
- Who it covers: any business using Facebook Pages, Instagram, WhatsApp, Messenger, Ads Manager or Meta Business Suite (PPC Land).
- Original wording: allowed "web scrapers, bots, robots, spiders, crawlers and user agents" on domains linked to the business. Content could be used to "provide, develop, improve, research, train and maintain the integrity of" Meta products (PPC Land).
- Revised wording, as reported: "train" is gone, and uses are tied to "advertising and business products" plus integrity and security (FourWeekMBA).
Technical analysis for researchers and developers
Meta's crawler documentation splits crawling by purpose: Robots.txt changes can take up to 24 hours to apply. The narrowed clause fits the Meta-ExternalAds use case (this is an inference). PPC Land notes the terms do not say robots.txt preferences take priority over the contract.
Read the full section
Meta's crawler documentation splits crawling by purpose:
- Meta-ExternalAgent: training foundation models; follows robots.txt.
- Meta-ExternalAds: ads and business-product improvement; follows robots.txt.
- Meta-WebIndexer: Meta AI search.
- FacebookExternalHit: link previews; may skip robots.txt for security checks.
- Meta-ExternalFetcher: fetches pages users request; may skip robots.txt.
Robots.txt changes can take up to 24 hours to apply. The narrowed clause fits the Meta-ExternalAds use case (this is an inference). PPC Land notes the terms do not say robots.txt preferences take priority over the contract.
Claims and evidence
- Legal experts: the original clause could permit scraping for AI training (Nieman Lab). This is legal interpretation, not evidence that scraping happened.
- Meta: crawling did not change, and the clause was always limited to ads and business products (TechCrunch).
- Meta: content owners can still use robots.txt (Yahoo/TechCrunch). Meta's own documentation is consistent with this.
Read the full section
- Legal experts: the original clause could permit scraping for AI training (Nieman Lab). This is legal interpretation, not evidence that scraping happened.
- Meta: crawling did not change, and the clause was always limited to ads and business products (TechCrunch). No independent audit of crawler logs backs this up.
- Meta: content owners can still use robots.txt (Yahoo/TechCrunch). Meta's own documentation is consistent with this.
- Exact revised wording: comes from secondary reporting (FourWeekMBA), not a published diff of the terms.
Context and prior work
- Meta* (June 25, 2025), Judge Chhabria found Meta's Llama training on books was fair use on that record.
- A separate case in Manhattan over book and journal piracy is still ongoing (Startup Fortune).
- On Dec 5, 2025, Meta signed undisclosed AI content deals with publishers including CNN, Fox News, USA Today, People Inc and Le Monde (MarketScreener).
Read the full section
- In Kadrey v. Meta (June 25, 2025), Judge Chhabria found Meta's Llama training on books was fair use on that record. He left room for future plaintiffs who show market harm (Perkins Coie).
- A separate case in Manhattan over book and journal piracy is still ongoing (Startup Fortune).
- On Dec 5, 2025, Meta signed undisclosed AI content deals with publishers including CNN, Fox News, USA Today, People Inc and Le Monde (MarketScreener).
Limitations, safety and contested findings
- Meta and the lawyers disagree on whether the original text allowed AI training. The disagreement was never tested in court.
- Even after the edit, the terms still authorize crawlers on domains linked to business accounts (FourWeekMBA).
- Two Meta crawlers, the link-preview crawler and the user-requested fetcher, may bypass robots.txt in some cases (Meta).
Read the full section
- Meta and the lawyers disagree on whether the original text allowed AI training. The disagreement was never tested in court.
- Even after the edit, the terms still authorize crawlers on domains linked to business accounts (FourWeekMBA).
- Two Meta crawlers, the link-preview crawler and the user-requested fetcher, may bypass robots.txt in some cases (Meta).
- Topic estimated the rights involved would likely cost many millions of dollars in a licensing market (PPC Land). That is an estimate, not a measured figure.
Business and practitioner implications
- Publishers: a training licence could have been granted through terms you accept just by continuing to use a product, which undercuts paid licensing deals.
- Site operators: block or allow each Meta crawler separately in robots.txt (meta-externalagent, meta-externalads, and so on) and check server logs.
- AI builders: public pushback reversed this clause within about two days.
Read the full section
- Publishers: a training licence could have been granted through terms you accept just by continuing to use a product, which undercuts paid licensing deals. Read platform terms updates the same way you read licensing contracts.
- Site operators: block or allow each Meta crawler separately in robots.txt (meta-externalagent, meta-externalads, and so on) and check server logs. Don't rely on contract wording alone.
- AI builders: public pushback reversed this clause within about two days. Getting training data through vague consent in terms of service carries real reputational and legal risk.