PolicyLaw, regulation & governance
Meta delays new business terms to November 20 after publishers object to website-crawling clause
Meta has moved its updated commercial and ad terms from October 30 to November 20, 2026. The clause letting it crawl websites linked to its business tools stayed in, and publishers say dropping the word 'train' does not bar AI training.
From the The Verge · AI article of October 23, 2026 and cited sources: Meta has postponed the terms, not withdrawn them. [1] [7] [2] [3]
SOURCES[1] The Verge theverge.com[7] Fast Company fastcompany.com[2] TNW: Meta delays crawling terms thenextweb.com
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- Original source
- The Verge · AI article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- October 23, 2026 article published
- Covered
- Oct 26 edition
- Evidence
- 11 sources across 10 sites
Reach: not measured yet
Meta has postponed the terms, not withdrawn them. They would let Meta crawl, index, cache and analyze any domain linked to a business that uses its ad, Facebook Pages or Instagram business tools. Businesses would accept simply by continuing to use those products, with no option to accept only part of the terms. The disputed wording was still in the terms after the delay was announced. [1] [7] [2] [3] [8]
Removing the word 'train' has not ended the dispute. Meta says crawling covers only ads and business products and that its crawlers will work the same way as before. Publishers and a Cornell law professor say the revised wording is still broad and does not rule out training. A UK-English preview reportedly still included the original sentence about AI training. [4] [8] [5]
robots.txt does not give site owners a clean way out. Meta-ExternalAgent reportedly uses a single token for both indexing and AI training, so a site cannot allow ad crawling while refusing training through that token. Blocking Meta's crawlers can break a publisher's Meta ads, and the terms do not appear to require Meta to follow robots.txt. [8] [2] [4]
Pressure on Meta is growing. The News/Media Alliance has demanded that Meta withdraw the terms. The UK Competition and Markets Authority has asked Meta to clarify them, and German publisher groups have asked the European Commission to review them. Meta is also being sued by publishers who allege it pirated books to train Llama. [3] [2] [6]
Meta calls the update a consolidation of permissions it already had, but no reviewed source independently compares the old and new terms.
Read the full assessment
If the clause takes effect, companies that advertise on Meta may effectively grant website access for AI uses through contract terms rather than licensing deals.
I'm writing up the dossier now that the sourcing is done.
Executive brief
Meta has pushed back its new commercial and ad terms from October 30 to November 20, 2026. Its statement says the extra time is for explaining "what's changing and what's not" (The Verge). The most important detail is what the delay leaves in place. As of the day after it was announced, the disputed crawling language was still in the terms (TNW/search summary). That language would let Meta crawl any website tied to a business using its ad, Page or Instagram business tools. Meta had already removed the word "train." Legal experts and publishers say that removal does not stop AI training, because the terms never expressly ban it.
What changed and event timeline
Publishers raise the alarm
The News/Media Alliance, which represents more than 2,000 US publishers, called the new terms "outrageous and probably illegal" ().
Meta drops "train."
Meta said crawling applies only to "ads and business products" and that how its crawlers work is not changing. The Alliance answered that training is no longer listed but is not banned either ().
Different versions in different regions
PPC Land reported that the UK-English preview of the terms still contained the original sentence about AI training. That leaves it unclear which wording applies to which accounts ().
Enforcement moved to Nov 20
Fast Company reported the delay first. Meta spokesperson Daniel Roberts pointed to robots.txt as the way site owners can signal their preferences (;).
Capabilities and access
No model is involved. The dispute is about access to websites.
Read the full section
No model is involved. The dispute is about access to websites.
- Who is covered: any domain or webpage linked to a business that uses Meta's advertising, Facebook Pages or Instagram business tools (TNW).
- What Meta may do: use "web scrapers, bots, robots, spiders, crawlers, and user agents" to "access, index, cache, analyze, collect, retrieve, or crawl" those sites, in order to "provide, develop, improve, research, and maintain advertising and business products" (Unite.ai).
- How consent works: businesses agree by continuing to use Meta's products. They cannot accept only part of the terms (PPC Land).
Technical analysis for researchers and developers
- robots.txt does not settle it. PPC Land reports that Meta-ExternalAgent uses a single robots.txt token for both indexing and AI training.
- Blocking has a cost. Publishers say blocking Meta's crawlers would break their ads, which depend on those crawlers (TNW).
- The terms may not bind Meta to robots.txt. The terms do not require Meta to follow it (TNW).
Read the full section
- robots.txt does not settle it. PPC Land reports that Meta-ExternalAgent uses a single robots.txt token for both indexing and AI training. A site therefore cannot allow ad crawling while refusing training through that token (PPC Land).
- Blocking has a cost. Publishers say blocking Meta's crawlers would break their ads, which depend on those crawlers (TNW).
- The terms may not bind Meta to robots.txt. The terms do not require Meta to follow it (TNW). PPC Land quotes Google's Gary Illyes describing robots.txt as an access preference that does not control what happens to data after it is crawled.
Claims and evidence
- Meta's position: the dispute is "largely a misunderstanding," and the update only consolidates permissions Meta already had (Social Media Today). This is Meta's own claim.
- Experts on the revised wording: Cornell law professor James Grimmelmann said it is still "broad enough to drive a truck through" (PPC Land).
- Publishers' money estimate: the Alliance says Meta owes US publishers about $2 billion a year (Unite.ai).
Read the full section
- Meta's position: the dispute is "largely a misunderstanding," and the update only consolidates permissions Meta already had (Social Media Today). This is Meta's own claim. No outside comparison of the old and new terms was found in the reviewed sources.
- Experts on the revised wording: Cornell law professor James Grimmelmann said it is still "broad enough to drive a truck through" (PPC Land).
- Publishers' money estimate: the Alliance says Meta owes US publishers about $2 billion a year (Unite.ai). This is an advocacy estimate that no independent source has checked.
Context and prior work
- Earlier fights with publishers: In 2022 Meta threatened to pull news from its platforms if the US passed the JCPA, a bill letting publishers bargain jointly with platforms (Inquirer).
- Training lawsuits: Five publishers are suing Meta, alleging it pirated books to train Llama (TNW).
- Regulators: The UK Competition and Markets Authority (CMA) asked Meta to clarify the terms (TNW).
Read the full section
- Earlier fights with publishers: In 2022 Meta threatened to pull news from its platforms if the US passed the JCPA, a bill letting publishers bargain jointly with platforms (Inquirer). Earlier in 2026 it lost an EU court case over payments to Italian publishers (BusinessDay).
- Training lawsuits: Five publishers are suing Meta, alleging it pirated books to train Llama (TNW).
- Regulators: The UK Competition and Markets Authority (CMA) asked Meta to clarify the terms (TNW). Germany's publisher associations BDZV and MVFP asked the European Commission to review them (MLex).
Limitations, safety and contested findings
- Wording not finalized: Meta has delayed the terms but not withdrawn them, and the disputed text was still there after the delay was announced (TNW/search summary).
- Version conflicts: different regional versions have shown different wording (ContentGrip).
- Training is still disputed. None of the reviewed sources shows how Meta actually uses data it has crawled.
Read the full section
- Wording not finalized: Meta has delayed the terms but not withdrawn them, and the disputed text was still there after the delay was announced (TNW/search summary).
- Version conflicts: different regional versions have shown different wording (ContentGrip).
- Training is still disputed. Meta says it is not changing how its crawlers work. Publishers point out that the terms do not exclude training. None of the reviewed sources shows how Meta actually uses data it has crawled.
- Reporting gaps: some secondary coverage gives conflicting names and titles for people quoted, so those quotes should be read with care.
Business and practitioner implications
- Publishers and brands that use Meta ads or Pages could find their whole website covered by the terms, not just what they post on Meta.
- Developers should check their logs for Meta crawler traffic and how it behaves. robots.txt rules alone may not cleanly separate ad crawling from training.
- AI teams should expect courts and regulators to look harder at training data obtained through contract terms rather than licensing deals.
Read the full section
- Publishers and brands that use Meta ads or Pages could find their whole website covered by the terms, not just what they post on Meta. Legal teams should compare the November 20 text with the current terms dated March 6, 2024 (Unite.ai).
- Developers should check their logs for Meta crawler traffic and how it behaves. robots.txt rules alone may not cleanly separate ad crawling from training.
- AI teams should expect courts and regulators to look harder at training data obtained through contract terms rather than licensing deals.