Meta removes AI training from disputed web crawling clause
October 18, 2026
After public criticism, Meta removed explicit permission to train on content from connected business websites from its new commercial terms. Broad access for advertising products remains.
What this is about
Meta changed its Commercial Terms scheduled to take effect on October 30, 2026 within one day. A version archived on October 16 explicitly allowed Meta to retrieve content from websites connected to a business account and use it to train Meta products. In the version visible on October 17, the word "train" and the corresponding sentence were gone.
The change followed sharp criticism from the News/Media Alliance. It matters to businesses, publishers, retailers and creators because linking a Meta business account to a website should not automatically be understood as consent to use all accessible content for model training.
What the changed clause actually does
The revised version still allows Meta to use automated software such as crawlers, bots and scrapers. These systems may access, index, cache, analyze and periodically update domains, webpages and URLs associated with a business user's products or services or made available through Meta products.
Meta now states that the purpose is to provide, develop, improve, research and maintain advertising and business products, as well as the security and integrity of its services. The archived earlier version included a separate sentence saying Meta could use website content to "research, train, and maintain" Meta products. That explicit training purpose has been removed.
This is a real textual change, but not a complete withdrawal of crawling rights. Meta may still technically collect connected pages and analyze them for broadly described advertising and business purposes.
Why it matters
The News/Media Alliance called the earlier language probably illegal on October 16 and warned that a business account could give Meta access to copyrighted material on an associated website. Meta spokesperson Andy Stone countered that the update was intended to pull accurate product information for ads and recommendations, not to scrape news sites. He also pointed to robots.txt preferences.
The sequence matters: Stone's explanation appeared while the explicit training language was still present in an archived version. A few hours later, that sentence had disappeared. Meta therefore changed the wording without publicly conceding that the original clause was too broad.
For affected businesses, the practical question is larger than the single word "train": What content is actually retrieved, how long is it stored, what derived data is created, and what choice do users have if they depend on Meta advertising? The published terms do not fully answer those questions.
In plain language
Imagine giving a delivery company a key to your shop so it can check prices for orders. The first contract also says it may copy everything inside to train its own systems. After protests, that sentence is removed. The key and permission to look around regularly still remain.
A practical example
A small German furniture brand runs an online store with 2,000 product pages and connects it to Meta Business Suite to place ads. Under the revised clause, Meta can periodically retrieve those pages, analyze product information and use it for advertising and business products.
The removed version would additionally have given explicit permission to use text and images from those domains to train Meta products. The new wording does not establish that this happens. Nor does the deletion alone prove that previously collected content is erased or that other Meta terms contain no comparable rights. The business should therefore review its account connections, robots.txt rules and the catalog data it actually shares.
Scope and limits
- The change concerns the wording of pending terms. It does not prove what data Meta has already collected or used internally.
- Removing the training sentence does not remove the broad permission to crawl, analyze and cache material for advertising and business products.
- robots.txt can express crawler preferences, but it is not a substitute for clear contractual limits and may not cover content supplied through other Meta features.
- No court has yet ruled on whether either version of the clause is enforceable in a particular country.
SEO & GEO keywords
Meta Commercial Terms, web crawling, AI training, Meta Business Suite, robots.txt, copyright, News/Media Alliance, business websites, advertising data, generative AI
💡 In plain English
Meta removed explicit permission to train AI on content from connected business websites. Permission to crawl and analyze those sites for advertising and business products remains.
Key Takeaways
- →The version archived on October 16 explicitly mentioned AI training.
- →The training sentence is absent from the version visible on October 17.
- →Meta retains broad rights to crawl, analyze and cache material.
- →Businesses should review connected domains, catalogs and robots.txt rules.
- →Actual data use and any deletion of collected material remain unclear.
FAQ
Did Meta stop crawling entirely?
No. The revised language still permits access, indexing, analysis and caching of connected websites for advertising and business products.
Can Meta no longer use the content for AI training?
The explicit training sentence was removed. That alone does not establish what rights other terms provide or how previously collected data is handled.
Is a robots.txt block enough?
It can express crawler preferences, but it is not a complete contractual or technical safeguard for every route through which content reaches Meta.
When are the new terms due to apply?
Meta lists October 30, 2026 as the planned effective date.