A customer used to find you, read your page, and call you. Some of them now ask an AI assistant to handle the whole errand instead. Find me a mortgage broker in Glendale and start the application. Book a pool cleaning for Saturday. Get someone out to look at the irrigation before the weekend.
The assistant goes looking. On most small business websites, it hits a wall.
This post is about that wall, why it is cheap to remove, and why I am not going to tell you it will make your phone ring next week.
What the Wall Actually Is
Put yourself in the assistant's position. It can read text and it can fill in a form. That is roughly the whole toolkit. Now look at a typical small business site:
- The contact form is a widget loaded by JavaScript from a platform vendor. The assistant cannot see the fields, so it cannot fill them.
- The booking button opens a calendar in a pop-up from a third party. Same problem.
- The phone number is the real way in, and the assistant cannot make a phone call.
- The robots.txt file blocks the crawlers the assistant's search tool relies on, usually because someone switched on a plugin to stop AI scraping.
So the assistant reports back that it could not finish the job. The customer does it themselves, or picks someone else. You never see any of it. No bounce, no form fill, no missed call. It does not show up in your analytics at all, which is the part that makes this easy to ignore.
Three Things That Remove It
None of these are projects. All three are an afternoon.
One: let the assistants read you. Open your robots.txt file and read it. If it blocks GPTBot, OAI-SearchBot, ClaudeBot or Claude-User, you have told the assistants to go away. There is a real argument for blocking crawlers that train on your content. There is no argument for blocking the ones trying to send you a customer. They are different crawlers with different names. You can allow one and refuse the other.
Two: say what you do in plain text. An assistant answering who does emergency drain cleaning in Pasadena on a Sunday needs four facts: what you do, where you do it, when you are open, and how to reach you. If those live inside an image, a PDF, or a video, they are invisible. Put them in text on the page. Add schema.org markup so there is no guessing. A plain llms.txt file listing your pages and services costs nothing and is read by the tools that are looking.
Three: make one path in that does not need a human. This is the one people skip. A plain HTML form with real field names, on your own domain, that emails you when someone submits it. Not a widget. Not an iframe. If an assistant can read the field names, it can finish the job for its customer, and you get a lead you would otherwise never have heard about.
What This Looks Like When You Go Further
We recently built the far end of this idea for a mortgage broker we work with. Not a contact form. The actual application.
A mortgage application is the Uniform Residential Loan Application, a federal form with roughly 200 fields. An assistant acting for a borrower can now send that entire form to the broker in one request. No account, no API key, no sales call first. A licensed loan officer reads it and calls the borrower back.
Three things made it work, and all three generalise to a much smaller business:
- One documented way in, described in a format machines read. The form is published as a schema, so an assistant can ask what the fields are instead of guessing at them.
- Honest failure. If the application cannot be delivered, the system says so and records nothing. It never reports success when there was none. An assistant that tells a borrower their application is filed, when it is sitting nowhere, has done real harm.
- A way to test that costs the business nothing. Anyone can run a complete application through it in preview mode. Nothing is sent, nobody is interrupted.
That last one is the lesson I did not expect. When you open a door for software, some of what walks through is a person checking whether the door is real. Build the door so that costs you nothing. Without a preview mode, the first week of attention would have buried a loan officer in applications from people who were only poking at it.
The Honest Part
Almost nobody is going to reach you this way yet. The volume today is small. I am not going to put a number on it, because any number I gave you would be made up.
So why do it?
Because the work is cheap, and it is not speculative. Plain text, real forms, an unblocked robots.txt and accurate schema.org markup are the same things that help you in ordinary search. You are not betting on assistants. You are doing standard, boring site hygiene that also happens to make you reachable if the behaviour keeps growing. If it never does, you still have a faster site, clearer content, and a form that works without JavaScript.
That is the whole case. Not a trend to get ahead of. A cheap option with a useful consolation prize.
What to Check This Week
Three things, in order of how likely they are to be wrong:
- Open yourdomain.com/robots.txt and read it. If you do not know what it says, that is the answer.
- Submit your own contact form from your phone. Did the email arrive? Check your spam folder. Then check the platform's spam folder, which is a different folder, and the one that usually has it.
- Ask an assistant what your business does, where you work, and when you are open. Whatever it says back is what your customers are being told.
The third one takes a minute and is usually the one that surprises people.