Open your analytics and look for the visits that did not happen. You can’t. That is the problem with this one.

When an AI agent shops for a buyer and your site turns it away, the buyer gets an error or an apology from their assistant, moves on to the next vendor, and you record zero. No bounce, no failed form, no lost-lead alert.

What Muse did to the picture

Meta launched Muse on September 8. It is a personal agent that runs on a cloud virtual machine with a real browser, and Meta says that when Muse browses, it “will appear as your activity.” It books travel, sends email, buys things.

Amazon blocked it on September 20. Its stated reasons: Meta gave no notice, the agent does not identify itself, and it appears to capture customer credentials. Meta replies that Muse has no visibility into passwords or payment methods. Shoppers saw a message saying that access by “an unauthorized AI agent” violates Amazon’s Conditions of Use.

Jessica Lessin, who founded The Information, wrote on X on October 4 that about half of what she used Muse for stopped working within two days, and asked whether sites were changing policy or detecting bots. That is one heavy user’s experience, not a measurement of the product. It does point the same way Amazon does.

TechCrunch’s survey of the problem is useful for what it shows about intent. Amazon blocked deliberately. Yelp wants paid access for agent traffic. United’s terms forbid automated tools without written permission. Walmart, a Muse partner, appears to have blocked it by accident, because its human-verification buttons fail when an agent touches them. Nobody there chose that.

Whether Muse announces itself is disputed. Fingerprint wrote on October 6 that it “provides no verifiable identity, alters its browser fingerprint, and suppresses the automation flags that most bot detection depends on.” A third-party crawler catalog lists a muse- token. I could not confirm either in Meta’s own documentation, so plan for the harder case.

The three places a site makes this call

robots.txt works only on visitors who say who they are. Terms of use work only if someone enforces them, which Amazon does. And the bot rules in your CDN or firewall run by themselves, at any hour, without telling anyone.

The third one wins. It was set by whoever runs your infrastructure or by the agency that built the site, and most marketing teams have never opened it. On September 15, Cloudflare changed its defaults: on pages with ads, the training and agent categories are blocked, and search stays open. If your domain is behind Cloudflare, find out whether that applies to you and who last looked.

Run this today

Skip the audit and do the test first, because it should take minutes and the result is hard to argue with.

Take an AI assistant that can browse. Ask it to find the price of your most-asked-about product and send a quote request. Write down where it stops. A verification widget, a form that fails, a pop-up, a login wall. Then ask IT or your agency three things: who last changed the bot settings, what blocked-request volume looked like in the last 30 days by category, and whether the price and quote pages are in that count.

After that, decide by page type. Reading prices and product pages can stay open. Buying, logging in, or collecting personal data can demand identification. That split is a business decision, and it is yours. Finally, get robots.txt, the CDN and your terms of use to say the same thing. They almost never do, because three different people wrote them in three different years.

The part that is not solved

Letting every agent in has a cost. An agent that hides its identity looks the same as one scraping your catalog or testing stolen passwords, and Amazon’s worry about credentials is not unreasonable. Blocking every agent has its own cost, a quiet one: the buyers who use assistants get a shortlist without you on it.

I do not know how many of your buyers use an agent today. Probably few. But when they do, nothing will announce that they stopped coming.

Where IQ Source comes in

The first hour of an AEO engagement with us is this list: who set the blocking, what an agent sees on your key pages, and what policy you want per category. Then it gets written into robots.txt, the CDN and the terms so they agree. I covered how to make a brand readable by an agent. This is the step before that, making sure the agent can get far enough to read it.

Do the assistant test today, on your own site. If it stalls somewhere, send me the step where it got stuck.

See what an AI agent sees on your site