Meta deleted the word "train" from its new terms. The crawl permission stayed.

What happened

On October 23, Meta said it would push the effective date of its updated commercial and self-serve advertising terms from October 30 to November 20. The company framed the delay as extra time for business users to understand "what's changing and what's not."

In the meantime, the draft itself changed. The original language said Meta's automated systems could "access, index, cache, analyze, collect, retrieve, or crawl" websites associated with a Meta Business account — and referred to using what it collected to provide, develop, improve, research and train Meta products.

Meta removed the word "train." It narrowed the stated purpose to advertising and business products. It did not remove the verbs that describe the crawling.

The word that got cut was the one that made headlines. The words that actually grant access are still in the document.

What it means

If you run ads on Meta, your site is probably "associated with" a Meta Business account. That association does not come only from an ad campaign. It can come from Business Manager, a pixel, a commerce catalog, a lead form integration, a domain verification record you added in 2021 and forgot.

So the operational question is not whether Meta is building a model on your archive. It's narrower and more answerable: what content can Meta collect from your site, for what stated purpose, and at what price. The answer to the third part is nothing. Meta has announced no compensation and no licensing program tied to these terms.

Publisher groups in the US and Europe have asked Meta to pause and define the scope. The delay to November 20 is the response so far.

What this is not

It is not confirmation that Meta is training a model on your website. The reporting does not establish that, and the explicit training language was pulled. If you say otherwise in a newsletter, you are ahead of the record.

It is not an opt-out. Nothing in the delay creates a toggle. Whether robots.txt, noarchive directives or crawler blocks override the contract language is not confirmed — contract permission and technical crawl controls are different layers, and nobody has published how they interact here.

It is not a news-publisher story only. Any ecommerce operator, agency, course seller or subscription site running Meta ads sits inside the same definition.

It is not resolved. A narrower stated purpose is not the same as a narrower grant. "Advertising and business products" is a category Meta defines.

And the November 20 date is not a negotiation window. There is no indication Meta is taking line-item feedback from individual advertisers.

The part worth watching

The interesting move is that Meta edited a contract in public, under pressure, in under a week. That tells you the terms were drafted without expecting anyone outside Meta to read them closely — and that reading them closely works. The same pressure has not produced a payment mechanism, which is the only thing that would change the economics.

What to do before November 20

1. Open Business Manager and list every domain verified or connected to your ad account. Include ones you inherited from a contractor.
2. Screenshot and save the current terms and the version taking effect November 20. Date the files. You want the before and after.
3. Email your Meta rep — in writing, not on a call — and ask one question: does this permission cover content collection for model training or AI answers, or only ad delivery and measurement?
4. Ask a second question in the same email: are robots.txt and crawler blocks honored under these terms, or does the contract supersede them?
5. Check whether your paywalled or member-only pages are reachable by any crawler that authenticates through an ad integration.
6. Where it's practical, keep Meta pixels and commerce tools off editorial subdomains. Separate infrastructure is the only control you fully own.
7. Do not publish that Meta is training on your site. Say what the terms permit, and say the training language was removed. That's the accurate version, and it's still a story.