A patient found our immigration exam clinic by asking Meta's Muse to do it. He never called and he never filled out a form -- the site answered his whole question before he showed up.
Shauntelle called me the other day to tell me a patient had found our clinic with the help of Muse. Which was funny, b/c I had told her about Muse the night before.
She runs WelcomeMed, our immigration exam clinic in Richardson. The patient told her he had asked Muse to find him an I-693 exam, and the agent had come back with us. A conversation with his Muse helped him find out exactly what it would cost, and told him that we would be his fastest route to his exam in a sealed envelope which would meet his requirement. He had not called anybody and he had not filled out a form. He just showed up excited and prepared.
Muse launched recently and we were on the tip of its tongue in less than three weeks.
Meta's agent runs in its own sandboxed browser and goes out to the actual web on your behalf. It is not searching in the sense we have meant for twenty years. It reads a site the way a careful person would, then reports back what it found.
So the only question that matters is what it finds when it gets to you.
Here is what it found with us. There is a file on our site named llms.txt that states in plain text every exam we offer and what each one costs -- adult, child, and both priority tiers -- along with what is included, who the civil surgeon is and where she trained, the address, the phone number, the booking link and the hours.
The page itself carries structured data saying this is a medical business, this is a physician, these are the hours, and here are the answers to the questions people actually ask. Our prices sit in the HTML as plain text, not behind a script that has to run first.
Our turnaround is stated honestly where a machine can read it. Same-day appointments, yes -- but documents in about a week, because labs take a week and no amount of wanting changes that.
Translation, we speak clearly in a language AI agents and LLMs understand.
An agent can answer a patient's whole concern fully from that without guessing anything. This one did.
The patient shared his conversation with me. This is how it started.
Then he told it he had no vaccination records. It did not send him away. It told him how the exam works without them.
I want to be careful not to oversell what happened. One patient is one patient, and I am not going to pretend three weeks of Muse is a trend. The honest version is that our site was ready for something that had not been invented when we built it, and the reason is that we stopped writing for a search engine and started writing for whoever needed the answer.
We especially got ready for AI agents.
Credit where it is owed, Shauntelle insisted the pricing be plain and complete long before any of this. She wanted a nervous patient to be able to read one page and know. The thing that makes a page good for a frightened person turns out to make it good for an agent, which is less of a coincidence than it sounds.
Business is going to be done by agents, for people. Not instead of people. On their behalf, with their money, at their instruction.
So go ask one. Open Muse or ChatGPT or Gemini, ask it to find and price what your company sells, and read what comes back as if you were the customer.
If the answer is vague, that is not the agent's failure. It is a description of your website's development roadmap.