Imagine someone walks into your store wearing your regular customer's jacket, your customer's shoes, even smelling like your customer's cologne. Nobody at the register questions it. That's basically what's happening on websites right now, except the visitor isn't a person at all.
It's an AI agent. And it's shopping with your money, on your customer's account, inside your customer's browser session.
Meta has one called Muse. OpenAI has one called Dots. You ask either of them to buy a shirt, book a flight, or reorder dog food, and they go do it. They browse, they fill in forms, they click "add to cart," they check out. No tabs, no comparison shopping, no typing in a credit card number at 11 p.m.
I've spent years chasing new ways to get around online — different booking tools, different platforms, different tricks for finding a flight or a flat that actually works. So when I read about this, my first question wasn't "is this good or bad." It was: what happens when the internet stops being built for people and starts being built for machines acting on people's behalf?
Turns out, that's already underway.
Two Ways In, and Most Sites Don't Know Which One They're Using
Muse gets into a business two ways. The easy way is a connector — basically an API handshake Meta builds directly with a company. Meta says it launched with "dozens of partners" plus the entire Shopify catalog, and it's adding Walmart, Best Buy, Sephora, and Wayfair.
The other way is the browser tier. If you haven't built a connector, Muse just shows up in a regular Chromium browser and works with whatever's on your page, the same way a person would.
Here's the part that caught me off guard: if you sell on Shopify, you might already be enrolled without doing anything. Shopify's default setting, tucked under Sales Channels, hands your product catalog to Meta automatically. A lot of store owners have no idea they've opted in.
No Name, No Rules
For twenty years, website owners have used a file called robots.txt to tell crawlers what they can and can't do. It works because crawlers identify themselves.
Muse and Dots don't. Meta's own documentation says when Muse visits a site, "it will appear as your activity." There's no name to block, because it's not claiming to be a bot — it's wearing your customer's identity.
Amazon didn't like that much. In September, Amazon blocked Muse entirely, not through robots.txt, but through its Terms of Use. GeekWire reported three reasons: Meta never told Amazon the agent would access its store, the agent doesn't identify itself, and it appeared to be capturing customer login credentials.
That's a company with enough leverage to say no. Most businesses don't have that option, or don't even know they need to decide.
Your Analytics Can't See the Sale
Here's where it gets genuinely strange. When an agent buys something through a connector, your website never sees the visit. The order shows up in your platform's records, not your analytics. Your traffic dashboard looks normal. Meanwhile, revenue is happening somewhere you're not watching.
Google Analytics added an "AI Assistant" channel in May to catch some of this, but it only works when the app sends a referrer. A lot of this traffic simply won't show up anywhere, at least not yet.
Why This Actually Matters to You
If you run a website, a shop, a booking page, anything where someone completes a transaction — this isn't theoretical. The agent reads your page through something called an accessibility tree, the same simplified structure screen readers use. If your "out of stock" label only shows up through styling, a gray button with a line through it, the agent can't see it. It has to click every option to find out, one by one.
Most SEO work — headings, alt text, visible facts — still matters, because the agent still has to find your site through a search. But once it arrives, the thing that decides whether it can actually finish the job is your page structure, your forms, your checkout flow. That's not an SEO team's job traditionally. It's about to become everyone's job.
Getting mentioned in an AI answer is nice. But if an agent can quote your website and still can't buy from it, you've basically become a nice-looking museum exhibit.
The web didn't wait for anyone's permission to change. It never has. The only real move now is to find out which tier you're already in, check what an agent actually sees when it lands on your page, and decide — on purpose — whether you want it walking through your front door.
Source: Search Engine Journal
Earlier in this series: Google's Search Traffic to Publishers Just Dropped 40%. Here's What That Means for Anyone Who Makes Things Online.
