Published September 29, 2026 · Last updated September 29, 2026 · Built on a contractor-authored field report from justforcontractors.ca (September 10, 2026) — the four-point scan is that author's proposal, not a tested universal method.

What Does ChatGPT Actually Understand About Your Business?

Direct answer: you have no idea — and looking at your own website can't tell you, because your eyes aren't the instrument the bot uses. A contractor-authored field report on justforcontractors.ca (September 10, 2026) started when somebody told the author they'd found his site, Just for Contractors, through ChatGPT. That one sentence pulled him into the bigger question: "What can AI search systems actually access and understand about your business?" His warning lands clean: "Your site can look perfectly normal to you and your customers, while something behind the scenes is making it harder for those systems to access or understand it." Before anyone sells you "AI optimization," he proposes a four-point pre-audit: (1) can AI get in, (2) can AI understand the business, (3) does the website give enough useful information, and (4) real buyer-question tests — built on the simpler question first: "Can they get in and understand what your business does?" Run that before you spend a dollar on anything with "AI" in the name.

Your eyes are the wrong instrument

Open your site in a browser. It loads, the phone number's there, the truck photos look good, the services are all listed. Done — looks fine to you.

That's the human experience. ChatGPT never has the human experience. It doesn't see your layout, admire your hero image, or click around the way a homeowner does. It reads what's retrievable and legible to a machine, and those two experiences can differ wildly. The justforcontractors.ca author put it this way: "Your site can look perfectly normal to you and your customers, while something behind the scenes is making it harder for those systems to access or understand it."

This is the pre-audit question almost no owner asks first, and it's written in contractor voice, not marketer voice — which is exactly why it cuts. A marketer sells you optimization. A contractor asks whether the machine can even get through the front door.

The question before the question

The field report's setup is worth quoting in full, because it names the trap: "Someone recently told me something that really got my attention. They found Just for Contractors through ChatGPT. … That got me asking a much bigger question about contractor websites: What can AI search systems actually access and understand about your business?"

Notice the order of operations. He didn't start with "how do I optimize for AI." He started with can it even see me, and if it can, what does it think I am? His framing sentence before any talk of optimization: "Before worrying about 'AI optimization,' I think there's a much simpler question to answer first: Can they get in and understand what your business does?"

That sentence is doing real work. Most of the AI-visibility conversation skips straight to tactics — schema, llms.txt, listing services — without ever checking whether the foundation exists. This is the foundation check. It costs nothing, needs no vendor, and takes an afternoon.

The four-point scan

The author's proposed pre-audit is four checks, in order:

Can AI get in? The access check. Is anything between your pages and the bots — blocking, challenges, technical walls? This is the binary gate: if the crawlers can't reach you, nothing downstream matters. Our sibling page covers the blocking audit in full — start with are AI bots blocked from your site — because that topic deserves its own treatment. Once you've confirmed access, come back; the remaining three checks are the ones nobody runs.
Can AI understand the business? The comprehension check. Not "can a human figure it out" — can the machine state what you do without guessing? The author's standard: "AI shouldn't have to guess what your business does." For a restoration contractor, that means: trade (water damage, fire/smoke, mold — which of them, specifically?), service area (which towns, named), and how a customer reaches you (phone number, emergency availability) — all of it present as plain readable text a machine can extract, not implied by a photo of a truck or a clever tagline. If the bot has to infer your trade from context clues, it will eventually infer wrong.
Does the website give enough useful information? The sufficiency check. A homepage that names the trade but never says which cities you cover, what happens on the first visit, or how fast you answer at 2 a.m. is thin for humans too — but a human fills the gaps with a phone call. A machine fills the gaps with a guess, or moves on to a competitor's site that doesn't make it guess. Sufficiency is judged against buyer questions, not against what your designer thinks a homepage should contain.
Real buyer-question tests. The evidence check. Ask ChatGPT the questions your customers actually ask — "who do I call for a flooded basement in Tacoma?" — and read what it says about your business. Not once as a vanity check: as a comprehension test. If it describes what you do accurately, checks 1–3 passed. If it guesses, hedges, or invents, checks 1–3 tell you exactly where to look.

The order matters. Testing (4) tells you something is wrong; access (1), understanding (2), and sufficiency (3) tell you where it's wrong. Owners usually run 4 first and panic. Run 1–3 first and you'll already know what 4 is going to say.

What "don't make it guess" looks like in restoration

Put yourself in the bot's seat for one page — your homepage. It reads the text. Now try to answer, from that text alone: What trades does this company do? Which towns does it serve? Is there an emergency number, and is it different from the office line? Are they actually licensed and local, or does the page just say "serving the greater area"?

Now picture the common failure modes. The trade lives in the logo and the photos, never in a sentence. The service area is "the Puget Sound region" — a phrase a machine can't turn into a town list. The phone number sits in an image or a click-to-call button with no text beside it. Hours are "call for a free estimate" instead of actual hours. Each of these is invisible to you, because you know your business and your brain fills the gaps before you notice them. The bot doesn't know your business. It only knows the page.

That's what "AI shouldn't have to guess what your business does" means as a working standard: every load-bearing fact about your business should be stated in plain words on the page, the way you'd say it to a stranger on the phone. If it requires interpretation, a human does it fine and a machine does it variably — which is the polite version of "sometimes wrong."

What to actually do this week

Ask ChatGPT what it knows about you. Type your business name and city, in plain language, and write down exactly what it claims — trade, area, phone, hours. Compare it line by line against reality. Every wrong or missing line is a gap in checks 2 or 3.
Confirm the access gate is open. Run the blocking check on our sibling page: are AI bots blocked from your site. Don't duplicate the work here — it's its own audit. If the gate is shut, fix that first; nothing else matters until it's open.
Read your homepage like a machine. Strip out everything that isn't text — ignore images, layout, and design — and answer the bot's-seat questions: trade, towns, phone, hours, emergency handling. Write down what you can't answer from text alone. That's your fix list.
Put the load-bearing facts in words. Trade, each town you serve, phone number, hours, emergency line — plain text on the homepage. Not in a graphic, not implied by a photo. Say it like you'd say it to a stranger on the phone.
Run one buyer question. "Who should I call for emergency water damage in [your town]?" Read the answer the way a customer would. If you're not named, or you're named with wrong details, you now know which of the four checks failed.

What to do this month

Run the full four-point scan and write it down. Date it. This is your baseline — the same way you'd photograph a job before demo. Without a baseline you can't tell whether anything you change later actually worked.
Re-test after every fix. Changed your homepage copy? Re-run the buyer-question test within a week. The scan is a loop, not a one-time event — which is why it pairs with a monthly AI-search visibility check.
Keep the scan vendor-free. No tested, independent comprehension-audit methodology is published anywhere yet. That means anyone selling you "AI optimization" today is selling confidence, not evidence. Run the four checks yourself first. If a vendor's pitch can't survive the four checks, it's not a pitch — it's a guess.
Watch the competitor question honestly. If ChatGPT names your competitor instead of you, run the four-point scan on their site too. Most of the time the answer is mundane: their site states the trade, the towns, and the phone number in plain text, and yours doesn't.

FAQ

Q: Isn't this the same as checking whether my site blocks AI crawlers?
A: No — that's check one of four. Blocking is binary: the crawler gets in or it doesn't. Comprehension is the other three: once it's in, can it understand what you do, and does your site give it enough to work with? The blocking audit is covered in depth on our sibling page (are AI bots blocked); this page is the comprehension pre-audit. Different job, same afternoon.

Q: My web designer says my site is fine. Why would ChatGPT see it differently?
A: Your designer checked the human experience — and for humans, it's probably fine. Nobody's disputing the design. The justforcontractors.ca author (September 10, 2026): "Your site can look perfectly normal to you and your customers, while something behind the scenes is making it harder for those systems to access or understand it." You need both checks: the designer's eye for humans, and the four-point scan for machines.

Q: Should I just buy an "AI optimization" package and skip this?
A: The contractor who framed this question says no — in his words: "Before worrying about 'AI optimization,' I think there's a much simpler question to answer first: Can they get in and understand what your business does?" The four checks cost nothing and tell you whether any package could even work. Optimization on top of a site the bot can't understand is paint on a house with no foundation.

Q: How do I actually test what ChatGPT understands about my business?
A: That's check four: real buyer-question tests. Ask it the questions your customers ask — your business by name, and your trade in your town — and read what it says about you like a stranger would. Accurate and complete means the first three checks passed. Guessing, hedging, or invented details means one of them failed, and the order of the scan tells you where to look first.

Q: Does this apply to Google's AI answers too, or just ChatGPT?
A: The author's question was about "AI search systems" generally — "What can AI search systems actually access and understand about your business?" The four checks are system-agnostic: access, understanding, sufficiency, testing. The comprehension problem is the same whichever engine you're checking — which is the point. Fix what the machines can't understand and you fix it everywhere at once.