AI can only answer what you have already written down
Every AI assistant setup page is really a form asking you to write down your business. Most owners have never done it, which is why the same question gets three different answers depending on who picks up. Here's the boring inventory that has to exist first — and why it's worth doing even if you never adopt a single tool.
Someone asks how far out she travels. She says twenty minutes, roughly. The website lists a set of towns she has not touched since she moved. Her husband, answering the phone last Thursday, told a caller yes to a job forty minutes out. All three answers are hers, and no two match.
Nobody is lying. There is simply no document. The true answer to how far she travels does exist — it lives in her head — but it comes out a little differently depending on who is asking, how full the schedule looks, and whether the job sounds worth the drive. On a slow stretch the radius grows. In a good stretch it shrinks. Her husband, doing her a favor by picking up, was working from a version he heard her give at dinner sometime last spring.
This is the unglamorous part of putting any assistant to work, software or human, and it is the part that never makes the brochure. An assistant does not know your business. It can only reflect back what already exists somewhere in writing. If the written version is thin, or scattered across a homepage, a voicemail greeting, and a text thread from March, that is exactly what comes back out — thin, scattered, and occasionally wrong in a way that sounds confident.
what the setup page is really asking for
Read the setup guides for the AI front-desk tools on the market right now and a pattern shows up fast. They all want the same short list: what you do, where you go, when you are open, how you take bookings, and the questions your phone gets over and over. Strip the branding off and the setup page is a form asking you to write down your business. The software is not the hard part. The form is.
The guidance for building any knowledge base a bot reads from tells the same story, and it is refreshingly dull. Audit what you already have. Throw out what is out of date. Standardize the words you use for things, because if you call the same service by three names, nothing downstream can tell they are the same service. And — this is the one everybody skips — resolve the contradictions before you load any of it in. If two of your documents disagree, the instruction is to decide which one is right. Not to hope the model works it out.
That last instruction is the whole game, and it explains why so many of these tools disappoint. The software cannot arbitrate between three plausible versions of an answer. Neither can a new employee. Neither, honestly, can you at four in the afternoon with a truck to unload.
the inventory
Here is the list. It is short, none of it is clever, and most of it can be drafted in an afternoon.
What you actually do. Not your trade — your services, named the way a customer would name them, not the way the industry does. A caller does not ask for hydro-jetting. They ask whether you can get a shower drain moving again. Write both, and write which of your services are routine, which need a look first, and which you will only take on if the job is already open.
What you deliberately do not do. This is the item everyone leaves out, and it is worth more than the list of services. Every business has jobs it turns away: too small, too far, wrong equipment, wrong insurance, wrong headache. Right now that boundary lives entirely in your judgment, which means it holds perfectly when you answer the phone and not at all when anyone else does. Written down, it becomes the thing that keeps the wrong job out of your calendar. Vendors of these tools quietly agree — their own documentation is full of scope controls and exclusion lists, because they know a confident assistant with no boundaries is worse than no assistant.
Where you go, and where you stop. Not a radius you feel. A radius you decided. Towns, zip codes, a highway you will not cross at rush hour, an exception for the good customer on the far side of the river. If there is an exception, say so — an honest rule with a written exception is usable. An unwritten instinct is not.
Your real hours versus your posted hours. These are almost never the same, and the gap is not a moral failing. Your site says you open at eight; you answer from six-thirty. Your site says closed Saturday; you take Saturday emergencies from people you know. Both facts are true. Only one is written anywhere, and it is the wrong one.
The handful of questions every caller asks. There are fewer than you think — how much, how soon, do you come out here, do you do this kind of job, are you licensed, do you take cards. You have answered each of them several hundred times and never once looked at how you answered. Write the answer you would want given if you were on vacation and someone else picked up.
That is the inventory. No tooling required. A text file is fine.
the contradictions are the point
Once it is on paper, the first thing you find is not a gap. It is a conflict.
This is where most owners get uncomfortable, and they shouldn't. The conflicts were always there — the document just stopped hiding them. Your site says one thing, your voicemail greeting says another, your listing on a directory you forgot you claimed says a third, and your spouse improvises a fourth. Each version was correct on the day it was made.
The contradictions have a cost long before any assistant enters the picture. Search engines pull your details from your site, your profile, and whatever directories have you listed, and when those sources disagree, the platforms can quietly reconcile the difference on their own — sometimes by publishing the version you least wanted. Customers experience it more directly. They show up when you are closed, or they call about a service you dropped, or they wait for a truck that was never coming out that far.
Fixing that is not an AI project. It is an afternoon with a list of every place your business information appears, and a decision about which answer is the real one.
it works with no AI in the room
Here is the part that makes this worth doing regardless of what you ever buy.
The document that makes an AI assistant useful is the same document a new hire needs on day one. The standard the operations people use is a good one: enough detail that someone who joined the team last Tuesday could handle the situation without interrupting anybody. That is a higher bar than most owners write to, and it is the right bar, because it is exactly the bar an assistant of any kind has to clear. Software is simply the least forgiving reader you will ever hand it to — it will not infer, it will not ask a colleague, and it will not notice that the answer stopped being true.
It is also what stops your website from contradicting your voicemail. That is worth something on its own, today, with nothing installed. A caller who gets two different answers from the same business inside a minute has learned something about that business, and it isn't about the answers.
And the inventory keeps working in every direction you might take this. Hand the front desk to a service, an answering company, a part-time hire, a spouse, or a piece of software, and every one of them needs the same page. The tool you pick is a preference. The page is a prerequisite.
For what it's worth, this is the business we're in at Catalyst, and it has made us blunt about the order of operations. The assistant on our side is called Sarah; she works over text today, not voice, and voice stays on the roadmap until it is genuinely there. Live texting for a business switches on as carrier approval for business texting clears, and until it clears, a person on our team covers the front desk by hand. None of that changes what has to happen first. Sarah, or any assistant like her, is only ever as good as the page she is reading — and that page is yours, not ours. Anyone selling you the reverse is selling you something.
the point
The interesting promise of an AI assistant is that it will know your business. It won't. It will know your notes.
That sounds like a limitation, and it is, but it is also the most useful thing anyone can tell you before you start: the work that makes the tool viable is work you can do this week with a blank page, and it holds its value whether or not you ever get around to the tool. Write down what you do, what you refuse, how far you go, when you are actually reachable, and the six questions you are tired of answering. Then go look at your website and see how much of it is still true.
Most owners who do this find the exercise humbling and then find it clarifying, in that order. The assistant is optional. The document was always going to be necessary.