When your next customer is an AI: staying visible on the agentic web

AI assistants are already choosing businesses on their customers’ behalf: comparing options, checking availability, and in some cases placing the order while the person who asked never opens a browser. An assistant can only choose a business it can actually connect to, which changes what “staying visible” means.

The front door is moving

The way customers find businesses has moved before. The phone book gave way to the website, the website to the search result, and the search result, in part, to the social feed. Each time the storefront stayed put while the front door moved, and the businesses that noticed early kept their customers.

It’s moving again. A customer can now tell an AI assistant, “find me a florist who can deliver peonies tomorrow before noon, and order them.” The assistant does the searching and comparing, and where it can, the buying. Your next visitor may be software acting for a person.

This stopped being fringe behavior a while ago. As of June 2026, Cloudflare reported that automated traffic had passed human traffic: 57.5% of all web requests now come from software rather than people clicking. One qualifier worth being honest about: that number counts requests, not human attention, and nobody is claiming most of your customers are robots. But a growing share of that automated traffic is assistants running errands for real customers with real money.

The loss is quiet

One in four business owners reports losing business because customers used AI tools instead (UPrinting survey, December 2025). The uncomfortable part is that this kind of loss doesn’t show up anywhere you’d normally look. No missed call, no abandoned cart. The customer asked their assistant, the assistant picked a business it could complete the task with, and you never found out the question was asked.

How an assistant decides who gets the business

An assistant acting for a customer has a job to finish, and it favors the path it can actually complete. To choose your business, it needs to do three things:

• Understand what you offer (services, products, prices, hours) from information it can actually read.
• Check the specifics: is the slot open Tuesday, is the part in stock, can you deliver to this address.
• Complete the action: book the appointment, place the order, request the quote.

A website only covers the first item, and only if the information is readable by software. From an assistant’s point of view, your website is a brochure. It can read the brochure, but it can’t transact through it. If a competitor offers a live connection with real availability and a booking the assistant can finish, while you offer a phone number and a contact form, the assistant takes the path it can finish. It’s the 9 p.m. customer choosing the restaurant with online ordering over the one whose line rings busy, happening at machine speed, all day.

That’s the rule worth writing down: AI assistants can only transact with businesses they can connect to.

What being “on the menu” requires

There’s now a standard way to offer that connection. It’s called MCP, short for Model Context Protocol, and it works like a universal connector between AI assistants and your business’s systems. It’s a vendor-neutral open standard, not owned by any single AI company, and ChatGPT, Gemini, Microsoft Copilot, and Claude all support it, so one connection covers the assistants your customers actually use. Stripe, Shopify, and PayPal already run official MCP servers. (New to the term? Start with What is MCP?)

In practical terms, getting on the menu takes three things:

• Information software can use: a clear, current description of what you sell (services, prices, availability), not a PDF, a photo of a menu, or “call for details.”
• At least one action an assistant can complete end to end: book, reserve, order, or get a real quote.
• Boundaries you choose: which requests get handled automatically, and which get handed to a human on your team.

Where to start

You don’t need to rebuild your business this month, and you shouldn’t try. Start by finding out where you stand: could an assistant, today, figure out what you offer and complete a simple transaction with you? For most small businesses the honest answer is “not yet.” That cuts both ways, because the menu is still short, and the businesses that show up on it early are easy to notice.

Readiness is more about process than technology. Our checklist, Is my business AI-ready?, walks through the five things to check, no tech degree required.

Last point. Handing the comparison-shopping and the 9 p.m. “are you open Tuesday?” traffic to software doesn’t remove people from your business. It means that by the time a human actually reaches you, the busywork is done, and your team has time for whatever it was that made customers pick you in the first place.