Meta’s New Commercial Terms Drop the Word ‘Train’: What Site Owners and Advertisers Should Do Before October 30

A single word, and why it mattered

Most terms-of-service updates pass without comment. The update to Meta’s commercial terms did not. In mid-October 2026, reporting and commentary spread quickly around one passage: language that appeared to give Meta broad permission to crawl websites connected to business accounts, and to use what it collected to train AI systems. Within days, by about October 17 according to a report attributed to Engadget, the word “train” had been removed from the revised terms preview.

That is an unusual sequence. Companies rarely edit legal text in response to a news cycle, and when they do, the edit tends to be narrow. Removing one verb changes how the sentence reads without necessarily changing what the company can do. That tension is the whole story, and it explains why the same episode is described as a win by some observers and as cosmetic by others.

The terms are scheduled to apply from October 30. For anyone who runs a website, buys Meta advertising, or publishes content that a business account links to, the practical question is not whether the wording looks better. It is what the document now permits, what it promises, and what you need to do before the date arrives.

What is actually confirmed

It helps to separate the reported facts from the interpretation layered on top of them, because the coverage mixes both.

• The effective date. The updated commercial terms were set to take effect on October 30. Coverage we reviewed did not confirm a change to that date.
• The original concern. Critics argued that the language could allow scraping of content from websites linked to business accounts, with AI training as a permitted use. Publishers were especially alarmed because they would otherwise be in a position to charge for continuous access to their articles for that purpose.
• The edit. By October 17 the word “train” no longer appeared in the relevant text of the revised terms preview.
• Meta’s position. Meta’s spokespeople have said the crawling terms apply to ads and business products, and that the edit does not reflect a shift in how Meta uses its web crawlers.
• The remaining criticism. Publishers say the revised language still does not rule out AI training, and that it does not promise that robots.txt will be honoured.

Several of the pieces circulating are from secondary aggregators rather than original reporting, and the original reporters’ articles were not directly reachable when we researched this. Treat any single outlet’s characterisation, including ours, as a summary to check against the actual terms text.

Why publishers care about crawling terms at all

To understand the reaction, it helps to understand the economics. For a news or reference publisher, a web crawler used for AI training is not a neutral visitor. It copies the product at scale, and the resulting model can then answer questions that would previously have sent a reader to the original page. Over the past few years, publishers have responded in three ways: blocking crawlers through robots.txt, negotiating licences with AI developers, and, in some cases, litigating.

Each of those responses depends on a clear line between permitted and unpermitted use. A licence has value only if there is something to license, and licensing revenue depends on the alternative, free access, not being available. Any term that quietly establishes a right to crawl for broad purposes weakens the position of every publisher at the negotiating table, even before a single page is fetched.

That is why the reaction was strong relative to the size of the edit. The concern was less about one crawler than about precedent: if a large platform can attach AI-use rights to a terms update that advertisers accept in order to keep running campaigns, the practical default for the open web shifts.

The robots.txt gap

The most concrete complaint after the edit is about robots.txt. This plain text file, placed at a site’s root, tells automated crawlers which parts of the site they may visit. It has no legal force on its own, but it is the near-universal mechanism by which site owners express preferences, and many AI developers publicly state that they honour it.

Publishers want a written commitment that Meta’s crawlers will respect it for these purposes. According to the reporting, the revised terms do not make that promise. Without it, a site owner’s only enforcement tools are technical, such as blocking by user agent or IP range, or legal and commercial, such as contract and licence terms. Those are costlier and slower than a one-line directive.

What Google Trends and the news feeds show

Search interest around AI and data terms has become a recurring feature of the technology news cycle, and this episode fits the pattern. On the day we compiled this article, technology feeds from the BBC, Ars Technica and TechCrunch were dominated by AI governance and corporate stories, and Ars Technica ran a headline stating that Meta had dropped the word “train” from its commercial terms after backlash. The Google Trends daily list for the United States, by contrast, was led by sport, weather and consumer queries, with no breakout term for this story.

That gap is informative. This is a story that matters intensely to a small, professional audience, publishers, advertisers, lawyers and platform policy teams, and barely registers with the general public. Its consequences, however, are likely to reach the public indirectly, through what AI systems know, which sites continue to publish freely, and how paywalls evolve.

What advertisers and site owners should do before October 30

You do not need to be a publisher for this to apply. Any business with a Meta business account that links to its own website is within the population the debate is about. A short checklist is reasonable.

1. Read the terms, not the coverage. Find the revised commercial terms and locate the sections on crawling, content access and data use. Note the exact wording, because this episode shows how much turns on single words.
2. Identify which sites your business accounts link to. The reported concern is about sites connected to business accounts. Make a list, including old campaign landing pages that nobody remembers owning.
3. Review robots.txt and server-side controls. Decide your policy on AI-related crawlers, apply it, and record the date. If you want a stronger guarantee than a text file, discuss user-agent or network-level blocking with your developers.
4. Ask legal counsel what acceptance means. Continuing to run ads after the effective date is typically treated as acceptance. If your content is a core asset, have counsel read the clause against your existing licences and any exclusivity commitments.
5. Assign an owner. Terms changes tend to fall between marketing, legal and engineering. Nominate one person to track this until the date has passed.

Reading Meta’s response charitably, and sceptically

There are two defensible readings of what Meta has said. In the charitable one, the original wording was broader than intended, the company clarified it when readers pointed out the ambiguity, and its statement that the terms concern ads and business products is simply accurate. Under that reading the edit is a correction and nothing else needs to change.

In the sceptical reading, a company that wanted to preserve optionality removed the most alarming word while keeping the structure that made the word possible. The fact that the revised text does not rule out AI training, and does not promise to honour robots.txt, is consistent with that view. One cannot distinguish the two readings from outside the company. What can be said is that a clear, written statement of what the crawling is for, and a clear commitment on opt-out, would resolve the dispute at almost no cost, and that the current text does not provide either, according to the criticism reported.

One report noted that the clarifying language on AI training applies only to ads. That is Meta’s framing, and it should be read as such rather than as settled fact. The distinction matters because “only for ads” and “never for AI training” are different promises, and the second is the one publishers are asking for.

The wider pattern: terms of service as policy

Zoom out and this is one of many episodes in which the rules governing AI and data are being set not by legislation but by contract updates. A platform changes its terms, users accept them, and a norm is created. Regulators and courts then arrive later, and the question becomes whether to undo a practice that is already widespread.

Kurums readers in procurement, legal and governance roles will recognise the pattern from our recent coverage of federal AI contract clauses and of export-control proposals: the substantive rules are increasingly found in the fine print, and the people most affected are often the last to read it. The remedy is boring and effective. Assign responsibility for monitoring terms changes from the vendors and platforms on which your business depends, and review them on a schedule, not when the headline arrives.

What to watch next

• The final text on October 30. Whether the live terms match the preview, and whether any further edits are made, is the first thing to check.
• Any statement on robots.txt. A public commitment would address the publishers’ main remaining objection.
• Industry and regulator reaction. Publisher groups may press for clarity, and regulators who are already examining how platforms use data may take an interest.
• Copycat terms. If other platforms adopt similar language, the practical default for the open web shifts further.

Bottom line

The removal of a single word is real, and it shows that public scrutiny can change a company’s drafting. It does not, on the reporting available, settle whether Meta may crawl business-linked sites for AI purposes, and it does not give site owners the opt-out guarantee they have asked for. Between now and October 30, the sensible course is to read the actual terms, document your own crawler policy, and ask counsel what acceptance commits you to. The date is fixed; the interpretation is not.

Sources: reporting from The Next Web, Slashdot, Startup Fortune, Ars Technica and other outlets between October 16 and 19, 2026; original reporting attributed to Engadget and Nieman Lab was not directly reviewed. Details may change; verify against Meta’s published terms.