Meta delays new business terms after publisher pushback on AI scraping

Meta has pushed enforcement of its new business terms from October 30th to November 20th. The terms drew fire because they appeared to let Meta scrape the websites tied to business accounts on its platforms and use that material for AI training. Meta already cut the word "training" from the text, but legal experts said that edit did not close the door.

The pressure was broad. The News/Media Alliance wrote to Mark Zuckerberg on Thursday warning of "severe harm" to publishers, and digital advertisers and publishers in the European Union and the UK asked regulators to force Meta to halt the terms. Meta's stated reason for the delay is more time to explain what is and isn't changing, including that site owners can still use robots.txt to signal their preferences.

We read this as a pause, not a retreat. Meta's own framing is that the delay exists so it can explain the terms better, which tells us the company still thinks the problem is communication. Publishers think the problem is the contract. Those are very different diagnoses, and only one of them gets fixed by a blog post and a few partner calls.

The robots.txt point is doing a lot of work in Meta's messaging. It is a long-standing, voluntary signal, and it has mostly held because big crawlers chose to honor it. If a company's commercial terms can be read as a separate grant of permission, the obvious worry is that the contract becomes the argument when the polite text file gets ignored. Meta says that isn't the intent. The fix for that worry is simple: put it in writing, in the terms, in plain language.

We're generally in favor of AI systems that learn from the open web, and we think broad access to information has real public value. But there is a line between reading the web and quietly converting every business that runs a Facebook page into a training-data donor through a clause most of them will never read. Consent buried in platform terms is not meaningful consent, especially for small publishers who depend on Meta for distribution and can't realistically walk away.

If this pattern scales, the web gets worse for everyone. Publishers lock content behind logins and blocks, smaller sites lose reach, and AI models end up trained on a thinner, more commercial slice of the internet. That is a bad trade for the very products Meta wants to build. I'm watching whether the November version comes back with tighter, defined language or the same terms and a nicer explainer.

The dispute began when Meta's updated terms for business accounts said it could use content from associated websites to develop, research and train its products. After backlash, Meta removed the training language, but lawyers quoted by Nieman Lab argued that broad words like "develop" and phrasing about content "associated with" a business's products could still cover scraping whole sites. The delay was first reported by Fast Company and confirmed to The Verge by a Meta spokesperson.