Your next customer is an AI agent

It will not browse your website. It will ask three questions about you, and if any answer is missing, it will pick someone else.

Last month a friend asked her assistant to find a physical therapist near her office, check who took her insurance, and book the first opening after 5 pm. It came back with two names and a confirmed appointment. She never saw a list, compared websites, or called anyone. The assistant handled all of it.

Somewhere in that city are forty physical therapists who would have been happy to take her. Thirty eight of them were not in the conversation. As far as they know, they lost nothing, because they never knew she existed.

Nothing in a report would show this loss. Traffic fades one conversation at a time, and the competitors who show up in the answer take the business.

What changed

A search engine showed ten options and let you fight for position. An AI assistant recommends one or two, and nobody hears about the rest.

For the last two years that was a visibility problem: are you mentioned when someone asks ChatGPT or Gemini or Perplexity a question in your category. It was serious, and most companies still have not addressed it. But it was still about a human reading an answer and deciding.

Now the assistants are turning into agents. Meta's Muse acts on your behalf. Grok Bot runs tasks on its own machine while you sleep. ChatGPT, Claude, Gemini, and Perplexity can all research, compare, and, increasingly, complete a purchase. The person states a goal and stops there, and the agent does the finding, evaluating, and buying.

An agent works from criteria rather than preferences, and it runs the same three tests on every company, in the same order, in about two seconds.

The three tests

Can I find you? The agent needs facts it can read without a human interpreting them. Your name, your locations, your hours, your services, your prices, in a form a machine can parse, and the same everywhere it looks. If your website says one thing, your listing says another, and a review site says a third, the agent does not resolve the conflict. It moves to a company whose facts agree.

Can I trust you? This is where most of the work is, and it is where twenty years of learning how search engines think transfers directly. Agents weigh the same signals a careful person would if they had unlimited time: real reviews, recent and specific; citations from sources the agent already trusts; structured data that says what you are in a language it understands; and content that answers the actual question a buyer asks, not the question your marketing team wished they asked. You do not need tricks for any of this, only a business a machine can read clearly.

Can I transact with you? This is the test almost everyone fails, and it is the one that decides the sale. The agent has found you and trusts you. Now it needs to book the appointment, get the quote, place the order, or hold the reservation. If the only way to do that is a form built for a human with a mouse, or a phone number that goes to voicemail after six, the agent cannot finish the job, and it completes the purchase with the competitor that gave it a door.

I call these Found, Chosen, Booked, and a business has to pass all three.

Why the third test is where the money is

Companies have spent two decades building for a person with a browser. Every page, every button, every checkout flow assumes eyes and a hand. Agents have neither. They need an interface, and unless you have built one for them, there is nothing for them to use.

There are two doors a business can open. The first is a machine door: live inventory and pricing an agent can query directly, and a checkout it can complete on behalf of its user. The standards for this are being written right now by the payment networks and the AI labs, and I have been in those rooms. The second is a voice door: a line an agent can hand a customer to, or call itself, when the conversation is what closes the sale. I built my first working version of that second door in 2019, three years before anyone was asking for it, and I have been building on it every day since.

Most businesses need both. A commerce company needs the machine door because the volume is there. A wellness practice, a home services company, an insurance agency needs the voice door because the sale still happens in a conversation, and the agent will happily start that conversation if you give it a number that answers. What that machine door looks like for an online store is laid out in AI agents for ecommerce.

What to do, in order

1. Fix the facts. Every place a machine can learn about you says the same thing. This is unglamorous and it comes first, because a single inconsistency costs you the trust test before you get to play.

2. Answer the real questions. Find the twenty questions a buyer asks an assistant on the way to choosing a company like yours, and make sure your site answers each one plainly, in text a machine can quote.

3. Open a door. That can be a machine door, a voice door, or both. The test is whether an agent can complete the transaction without a human on your side touching it. If it cannot, the agent will shop somewhere else.

4. Measure the phrases that produce revenue. Skip the vanity phrases and track the ones that end in a booking, an order, or an appointment. Run them across the major assistants every month and watch whether you are the answer.

The timing objection

The most common thing I hear from executives is a version of "we will move when it settles." I understand the instinct, and I have been on the wrong side of it in the other direction. In 2019 I built something the market was not ready to buy, and I learned that seeing what is coming is much easier than getting the timing right.

The timing matters now because agents are forming defaults. When an assistant finds a company that passes all three tests, it remembers and goes back. The businesses that are legible and transactable in the next eighteen months become the answer, and the answer is sticky in a way a search ranking never was.

You do not need to rebuild everything. You need to pass the three tests, and you need to pass them before your competitor does.

The assistants are already making these choices. It is worth finding out whether they are choosing you.