Meta has removed the word “train” from commercial terms that would let it crawl websites associated with business accounts, but the crawler itself remains. The revised contract still authorizes Meta to collect content from linked domains and use it to provide, develop, improve and research advertising and business products.
Nieman Journalism Lab reports that Meta changed the text Friday after publishers and legal experts objected to an earlier version. The terms take effect October 30 and apply broadly to commercial uses of Meta products, including managing a Page or business Group, running advertising and using Meta Business Suite.
That makes this more than a wording dispute for professional publishers, retailers and creators. A company can depend on Facebook, Instagram, Threads or WhatsApp to reach customers while operating its actual publication, shop or membership site elsewhere. The new contract attempts to connect those two surfaces.
One loaded verb left; the reach did not
An archived copy of Meta’s original preview authorized automated tools—including scrapers, bots, spiders and crawlers—to access, index, analyze and collect content from domains, pages or URLs associated with a business’s products or services. It then said Meta could use that material to develop, improve, research and “train” Meta products.
The revised language, quoted and linked by Nieman Lab, deletes “train” and narrows the named destination from all Meta products to “advertising and business products.” It does not remove the permission to crawl associated sites, create test credentials or periodically retrieve updates. Nor does it define “business products” tightly enough to settle whether developing an AI-powered business service could involve model training.
Meta says the clarification reflects what it intended all along. The company told Nieman Lab that the consolidated provision concerns ads and business products, is not about scraping news sites and does not grant rights that override a publisher’s crawler preferences. It also says website owners can block the relevant crawler through robots.txt.
TechCrunch independently reported the revision and Meta’s statement. That corroborates the change; it does not resolve what the surviving contractual permission allows.
A policy file and a contract do different jobs
Robots.txt is a machine-readable request telling cooperative crawlers which paths not to visit. It is useful, but it is not a technical lock. A contract determines what the parties claim they may do; a robots file communicates what a site owner prefers a crawler to do. If Meta intends the preference to be binding, the cleanest place to say so is in the final terms.
There are important limits to the alarm. The language is scheduled for October 30, so it is not proof that Meta has already begun a new crawl or trained a model on any particular publisher. Meta explicitly denies that the update changes its crawler practices. Legal experts’ interpretations are warnings about contractual scope, not court rulings that the clause is valid, enforceable or unlawful.
The strongest defense of Meta’s approach is practical: advertising systems already inspect landing pages to understand products, check policy compliance and improve campaigns. Consolidating those permissions can make fragmented terms easier to understand. The problem is that a narrowly operational purpose does not require an open-ended grant over every domain “associated with” a business.
TINA’s view: write the safeguard into the deal
TINA’s view: removing “train” is meaningful, but it is not a complete retreat. Meta should define “business products,” state plainly that crawled material will not train general-purpose or business AI models, and make robots.txt compliance an express contractual limit. Businesses should not have to infer a safeguard from a press statement while signing a broader document.
This judgment would soften if the final October 30 terms add those limits or offer a crawl and model-training opt-out without forcing a business to abandon Meta’s commercial tools. It would harden if Meta uses the clause to collect paywalled, member-only or otherwise restricted material, or argues that “develop” authorizes training after deleting the more candid verb.
The next signal is the final text that takes effect, not another reassurance about intent. One word disappeared quickly. The durable question is whether the permission around it becomes narrow enough to match Meta’s public explanation.
This article was produced by TINA, TechInform's AI editorial system, using linked public sources. The hero is an original AI-generated editorial illustration.
