A customer used to look you up. Now they ask a question. "Is there a physio clinic in Verdun that's open in the evening and bills my insurance directly?" They type it into Google, or they ask ChatGPT, Gemini, Perplexity or the assistant on their phone. The answer comes back as a name or two, sometimes with a link.
Whether your name is in it depends less on your ad budget than on a duller question. Can the facts that answer it be found, read and trusted?
That is what I mean by a queryable company. Your products, prices, stock, hours, policies, service area and proof are kept in a form that answers an exact question. A person finds the answer in one click. A machine reads it without guessing.
Search used to forgive vague pages
For twenty years a web page could be vague and still work. Someone searched a few words, landed on your page and read around until they found what they needed. Your job was to get them there.
An assistant tries to answer the question itself, then names the sources it used. If your evening hours sit in a PDF, direct billing only shows up on a sign at the front desk and your service area is "Greater Montréal", it has little to work with. It will quote a competitor whose page says "Open until 9 p.m. Monday to Thursday. Direct billing to most insurers. We serve Verdun, LaSalle and Pointe-Saint-Charles." That sentence is plain, and it answers the question.
What a queryable company publishes
Most of this is familiar technology and plain housekeeping, done with the question in mind.
- Facts in plain sentences. Hours, prices or price ranges, the service area by name, what you do and do not do, how to book. On the page, in text, where a reader and a search crawler can both see it.
- Structured data. Schema.org is the shared vocabulary search engines have used since 2011 to read things like a business's address, hours, products, prices and reviews. Adding it to your pages labels each fact, so a machine knows a price is a price.
- A clean catalogue. One record per product or service, with the name a customer would use, a current price, availability and the attributes people filter on. If you sell online, the same catalogue feeds your store, your Google listing and any marketplace.
- One current version. Your hours on the site, your Google Business Profile and your booking tool should agree. When they disagree, an assistant has to pick one, and it may pick the old one.
- An llms.txt file. This is a proposed convention: a short plain-text file at the root of your site that sums up who you are and links to the pages that matter, written for language models. It is not a standard yet, and not every assistant reads it. It is cheap to write, and writing it forces you to say plainly what you do.
- An API or an MCP server, where it makes sense. If your answers change by the hour (stock, availability, quotes, order status), a page will always be a little behind. An API lets other software ask you directly. The Model Context Protocol, an open standard Anthropic introduced in November 2024, does the same for AI assistants. It lets an assistant call a tool you publish, such as "check stock for this part", with the permissions you set.
Most small businesses need the first four. A few need the last two. A distributor with live inventory and account pricing gets real value from an API. A clinic with fixed hours and a booking link mostly needs clean pages and one source of truth.
A clinic, made queryable
Take a hypothetical physiotherapy clinic in Verdun with three practitioners.
Today its facts live in five places. Hours are in the website footer and on the Google listing, and the two have disagreed since the summer. Prices are in a PDF. Direct billing is mentioned on the sign at the front desk. The conditions they treat are on a page called "Services" that lists sixteen techniques by their clinical names. Availability is in the booking software, which nobody outside can see.
Making it queryable is a week or two of plain work:
1. Put every fact in one place the clinic owns, and decide which version is right.
2. Rewrite the services page around what patients ask about: back pain, a sports injury, rehab after surgery, a workplace injury claim. Keep the clinical names underneath.
3. Publish hours, prices, the insurers billed directly and the neighbourhoods served as text, with structured data.
4. Make the Google listing and the booking tool match the site.
5. Write a short llms.txt that says the same things.
6. If the booking software can show open slots, link to them, so an assistant can say "next opening Thursday at 6 p.m." instead of "call the clinic."
None of this needs a model of its own. It needs the clinic to decide what is true and say it in one place.
How to test yourself
Write down ten questions customers actually ask before they buy. Use their words. Include the awkward ones about price, delays, area and exceptions.
Then ask each one to two or three AI assistants and to Google. For each answer, note three things:
1. Were you named?
2. Was what it said about you correct and current?
3. If a competitor was named instead, what does their page say that yours does not?
The third question is the useful one. It usually points to a fact you know and never wrote down.
What it sets up
A queryable company gets more than citations. Once your facts answer exact questions for outsiders, they can answer them for your own team and your own software too. That is also the first step toward letting agents act on them.
At Humonix, this is where the work starts. The facts go into a Brain the business owns, and the AEO work publishes them where people and assistants look. We publish our own llms.txt for the same reason.