Found. Chosen. Booked.

Every AI agent that shops on your customer's behalf runs three tests on your business before it completes the transaction. Most companies fail at least one.

An agent is shopping for your customer right now, with a goal, a budget, and instructions. It will evaluate 50 companies in three seconds and return two names. One of the two gets the business, and the other 48 never learn the customer existed.

The agent does not care about your brand, your testimonials, or how much you spent on your website. It runs three tests in order, and if you fail any one of them, it moves to the next company. Most businesses pass the first test, fail the second, and never get to the third. When a pipeline thins out for no obvious reason, this is often why.

The three tests

The agent needs your name, category, location, hours, services, and pricing in a form a machine can read. That means structured facts it can reach without a browser, consistent across every place the agent might look, rather than marketing language, a PDF, or a JavaScript dropdown.

Your website says you are open Monday through Friday 9 to 5. Your Google profile says Tuesday through Saturday 10 to 6. A directory listing from 2019 says you closed. The agent will not sort this out. It sees the conflict, marks you unverifiable, and moves to a company whose facts agree.

Companies tend to assume they pass this test, and most do not. Having a website and a listing does not make you findable. To an agent, findable means your facts are machine readable and identical everywhere they appear. Most companies fail on data alone, before the agent even evaluates quality.

Why most businesses fail Test One

Your facts live in 40 places, were entered by 12 different people over six years, and nobody has audited them since. Your hours changed during COVID and you updated your website but not your Google profile. Your services expanded but your Yelp page still lists what you offered in 2020. Your address is formatted three different ways depending on who submitted it.

A person sees this and makes allowances. An agent sees it and assigns you a trust penalty. One inconsistency costs you points, and three can cost you the sale. The agent has 20 other companies with clean data, so it has no reason to give you the benefit of the doubt.

Most companies spend six figures on their website and zero hours making sure the 40 other places a machine reads about them say the same thing.

Say the agent has found five companies with consistent facts. Next it evaluates trust. This is where SEO and content strategy transfer directly, and it is where most of the competitive work happens.

The agent is looking for recent reviews that mention specifics, citations from sources it already trusts, structured data that declares what you do in a vocabulary it understands, and content that answers the actual question the buyer asked, not the one your marketing team wanted them to ask.

There is no trick to it. The agent weighs the same information a diligent person would if they had unlimited time, except the agent really does have that time, and it checks everything.

Why passing Test One and failing Test Two is the most common mistake

Companies that fix their data tend to think the work is done. Their facts are clean, so they assume the agent will choose them, but clean facts only get you into the evaluation set, and trust decides who wins it.

Your competitor has 84 reviews in the last six months. You have 12 in the last two years. Your competitor is cited by three local news sites and a trade publication the agent recognizes. You are cited by a directory scraper and a link farm. Your competitor has schema markup that tells the agent exactly what services are offered, at what price, with what guarantees. You have paragraphs of prose the agent cannot parse.

The agent weighs all of this in 200 milliseconds and picks your competitor. You never see the traffic or learn you were in the consideration set, so it looks like demand is down, when the agent is really sending that demand to someone else.

This is where companies that understand SEO have a structural advantage. Twenty years of learning how to be legible to Google transfers directly to being legible to agents. Structured data, citations, and content that matches query intent matter to agents the same way they matter to Google. If you have been doing this work, Test Two will look familiar, and if you have not, you are behind the companies that have.

The agent has found you and chosen you. Now it needs to finish the job. It needs to check availability, get a quote, book the appointment, place the order, or hold the reservation. If the only way to do that is a contact form that goes to an inbox someone checks twice a day, or a phone number that rings through to voicemail after six, the agent cannot complete the booking, so it goes back to the company that came in second and tries there.

This is the test that decides the sale, and it is the one almost nobody is ready for.

What Test Three requires technically

There are two doors you can open. The first is a machine door. The agent queries your availability, pulls live pricing, and completes the checkout on behalf of its user. This requires an API or a structured integration the agent can call without human intervention. The standards are being written right now by the payment networks and the AI labs. Companies that sell products, hold inventory, or offer scheduled services at published rates need this door.

The second is a voice door. The agent hands the customer to you at the right moment in the conversation, or it calls on the customer's behalf when the transaction requires explanation. This is what works for businesses where the sale happens through consultation: wellness practices, professional services, home services, and anything with complexity or customization.

Most businesses need one of these doors and many need both, yet almost nobody has either.

Your website and booking flow were built for a person with eyes and a mouse, and your checkout has six steps and a CAPTCHA. None of that works when the buyer is a machine acting on instructions. You do not need to rebuild everything. You need one door an agent can walk through, and you need it before your competitor builds theirs.

How to run the tests on your own business

Write down the 20 questions a buyer asks when deciding between you and a competitor. Run them through ChatGPT, Claude, Gemini, and Perplexity. See where you appear, where you do not, and what the agent says about you when it does mention you.

Check the facts. Google yourself on every platform an agent might read. Do your hours match? Does your address? Do your services? If you find one inconsistency, assume there are 10 more you have not found yet.

Try to book yourself. Pretend you are the agent. Start from scratch with nothing but your company name and the city. Can you find availability without calling? Can you get a quote without filling out a form? Can you complete the transaction end to end without a human approving something? If not, neither can the agent.

The companies that run these tests in the next 90 days and fix what they find will be the defaults agents recommend for the next five years, and the ones that wait will spend those years buying the position back.