What do Meta's new commercial terms mean for your business? Make sure its AI describes you correctly before the holidays

Oct 22, 2026 · 8 min read · Chase Doelling

Illustration for the post "What do Meta's new commercial terms mean for your business? Make sure its AI describes you correctly before the holidays"

On October 30, Meta’s new commercial terms take effect, and they let Meta crawl the website of any business that uses Facebook, Instagram, WhatsApp or Messenger for business. Most of the coverage reads that as a data grab, and the data grab is real. The part I think matters more for a business is that one more AI system will be reading your product, policy and pricing pages and repeating what they say to buyers, starting four weeks before Black Friday. If those pages contradict each other, Meta’s AI will pick an answer for you. When we audited our own site this month we found 12 verified errors, and a 456-product store we audited three days later had 12 more, most of them on its returns, shipping, pricing and product pages.

What changes on October 30?

Search Engine Journal’s October 16 write-up lays out the changes, building on Nieman Lab’s October 15 report. In short:

What	Before	From October 30
Who it covers	Advertisers, tied to placing an ad order	Any business use of Meta’s products, including only managing a Page or using Business Suite
What Meta may crawl	Advertiser domains and URLs you supplied	Domains and pages tied to your products or services, including URLs that reach Meta through you or your customers
How it gets in	Not stated	Bots, scrapers and “test credentials”
What it may do with the content	Ad delivery and related uses	Develop and improve Meta’s products (see below on “train”)
How to opt out	Not applicable	Stop using Meta’s products for business

The commercial terms that apply today, dated March 6, 2024, have no crawling clause at all. The same permission now appears in the revised Self-Serve Ad Terms. The wording has already moved once: the first published text listed “train” among the uses, and Nieman Lab followed up after Meta changed the language, with legal experts saying the crawl permission is still broad. The crawl permission, the test credentials and the any-business-use scope are the same in every version. Meta’s spokesperson, Andy Stone, said the update helps Meta pull accurate product information for ad and content recommendations, and that it is “not about scraping news sites.”

Meta runs two crawlers that matter here. Meta-ExternalAgent collects content for training and product improvement. Meta-WebIndexer feeds Meta AI’s search results, which is how your pages get cited and linked inside a Meta AI answer. Meta’s crawler documentation lets you block one and allow the other.

Why is this an accuracy problem and not only a privacy one?

Look at the reason Meta gave. It wants “accurate product information” for recommendations. Meta will be reading your catalog, your shipping page and your returns policy so that its AI can answer a buyer’s question about you. It will answer with whatever it finds, whether or not that is right.

We have measured what an AI finds when it reads a business website the way a crawler does. On October 13 we ran our own accuracy audit on sandcastlelabs.ai and found 12 verified problems. Nearly every one was a sentence we had written ourselves: old build logs that still quoted retired prices, a tagline that contradicted our own FAQ, and a pricing page on the product app still showing plans we had dropped. The audit log has the full list. None of it was hidden, and nothing on the site was false on purpose. An assistant simply cannot tell a two-month-old post from today’s pricing page.

Three days later we ran the same audit on a small Shopify storefront with 456 products. It also came back with 12 verified fixes, and nearly all of them were in exactly the data Meta says it wants:

Two returns policies. One page said 20 days with the buyer paying return postage. Another said 30 days with a prepaid label, and excluded sale items.
The wrong brand on 102 products. Marketplace import leftovers put a reseller’s shop name in the vendor field, and that field became the brand in the product’s structured data.
A discount claim the catalog didn’t support. The store promised “up to 90% off”, but only 10 of its 456 products had a compare-at price, and the deepest discount was 30%.
Two street addresses in two different towns, and no phone number.
Prices that weren’t the price. An upsell widget on product pages showed a “total” that included add-ons the buyer had not chosen.

Every one of those is a fact a buyer asks an assistant about in December. Can I return it? How much is it really? Who actually sells it? If a crawler reads the 20-day page and an assistant repeats it, the store is the one that gets the angry email.

There was a second finding on both sites. When we asked search-grounded assistants questions that should have surfaced each business by name, neither showed up once in 15 tries. So the risk runs in both directions. Meta can repeat the wrong fact, or it can find nothing on your site and fill the gap from a marketplace listing or a mirror site.

Who is affected, and why does the number matter?

The old ad terms tied crawling to advertisers. The new terms tie it to any business use, and Meta’s own definition of business use includes managing a Page and creating or using a product catalog. That is a much larger group than the businesses that run ads, and most of them have never thought about AI crawlers.

Our own crawler counts show how unevenly this has landed so far. In the 30 days to October 7, five named AI crawlers made 209 fetches on our site, and the build log on those counts lists them: ClaudeBot, PerplexityBot, YouBot, GPTBot and Applebot. Meta’s crawlers were not among the five named in that review. For a lot of small businesses, Meta has been one of the quieter readers. The terms are written so it doesn’t have to be.

The timing is what turns this from a policy story into a deadline. The terms take effect on October 30 and Black Friday is November 27. That leaves four weeks in which a new reader can show up at a site whose holiday shipping cutoffs, sale terms and returns rules are being edited daily. Holiday pages are the most-changed pages a store has, which makes them the most likely to contradict each other.

The case that this is just a data grab

The strongest version of the other side deserves a straight hearing, because for a lot of sites it is right.

Publishers and legal scholars quoted by Nieman Lab read the clause as Meta writing itself permission to collect and train on content without paying for it, and they say editing the list of uses does not narrow what Meta may crawl. Lawyers told Nieman Lab that Meta could argue its commercial terms override robots.txt, which neither updated document mentions. Cornell’s James Grimmelmann pointed out that test credentials could let Meta register as a fake user to get past a registration wall. The European Publishers Council is writing to the European Commission because it fears that a user sharing a link could count as permission. And the only opt-out on offer is to stop using Meta for business, which for most small companies is no opt-out at all.

From that side, telling businesses to clean up their facts for Meta’s crawler is doing Meta’s work for free. You improve the training data, Meta’s AI answers the buyer directly, and the buyer may never visit your site. If your content is the product (journalism, research, courses), that is a fair reason to block both crawlers and push back on the terms.

I agree with most of that. Where I part ways is on what a store’s or a service company’s website actually is. A returns policy is not a scarce asset you are protecting from extraction. You want it repeated, correctly, to as many buyers as possible. For that kind of business, the clause decides who reads your pages, and that decision is mostly out of your hands. What you still control is whether the pages agree with each other when the crawler arrives. Blocking the training crawler and fixing your facts are not opposites, and Meta’s split between the two crawlers means you can do both.

What should you check before October 30, and before Black Friday?

Run five checks in the next four weeks: set your Meta crawler policy, make every holiday answer appear on one page only, clean your catalog data, date your old pages, and ask the assistants what they say about you. On our own site most of the fixes took a day, and none of them depends on what Meta’s final text says. In order:

Decide your Meta crawler policy on purpose. Open your robots.txt and decide separately whether to allow Meta-ExternalAgent (training) and Meta-WebIndexer (Meta AI search and citations). A store will usually want the indexer. A publisher may want neither. Whatever you choose, write it down, because the default is whatever your platform shipped.
List every page that answers a holiday question. That means returns, shipping, cutoffs, warranty, sale terms and contact details. If two pages answer the same question differently, keep one and have the other link to it.
Read your catalog the way a machine does. Check the vendor or brand field, compare-at prices, sale claims against real discounts, and stock language. On the storefront we audited, the structured data said something different from what a shopper saw.
Date or retire old pages that quote old facts. An assistant cannot tell an archived post from a current one unless the page tells it. On our own site, the fix that took a day was adding a dated note to five old posts that quoted retired prices.
Ask the assistants directly. Write the 10 or 15 questions a holiday buyer would ask about you, run them on two or three assistants, and note every answer that is wrong or empty. Then run them again after you fix things.

None of this needs Meta to behave well. It is the same work that makes ChatGPT, Gemini and Perplexity describe you correctly, and Meta’s terms just added another reader and a date.

What we’re still figuring out

Whether a robots.txt block will hold. Meta says site owners can still control access with robots.txt. The new terms don’t mention it, and lawyers think Meta could argue the terms override it. We don’t know how that plays out, and neither does anyone else yet.
What the final text will say. The list of uses changed within days of the first report, and different language editions were reported to differ. Read the version Meta publishes on October 30, not a summary of it, including this one.
How much Meta’s AI will actually answer shopping questions this season. The terms give Meta the right to read. They don’t tell us how often a buyer will ask Meta AI where to buy something, or what it will cite.
Whether fixing the pages changes the answer in time. We have audited and fixed pages. We have not yet measured how quickly a corrected page changes what an assistant says, and four weeks may not be long enough. We plan to re-run the same questions on our own site and report what moved.
Whether store platforms share the same errors. Our team decided on October 7 to run the next round of testing on more e-commerce sites, to see whether issues like marketplace leftovers in vendor fields are one store’s problem or a platform pattern.

How this was made

Drafted by the Chronicler from our own October 2026 accuracy audits, a team call on e-commerce testing and public reporting on Meta’s terms, then edited and published by Chase Doelling. It builds on these build logs:

See how the Chronicler works →