Is ChatGPT recommending your business? Here's the 30-minute check

You can check your Google ranking in about five seconds. Type the search, count down the page, there you are — or there you aren’t.

Now answer this one: when someone asks ChatGPT “who’s the best plumber near me,” does your name come up?

Most business owners have no idea. Not a vague idea — no idea. And that’s a strange blind spot, because a growing share of buying research now starts in a chat window that never shows a list of ten blue links. There’s one answer, it names two or three businesses, and everyone else is invisible.

Here’s how to find out where you stand. It takes about thirty minutes and costs nothing.

Step 1: ask the way a customer asks

The most common mistake is testing your own brand name. You type “what do you know about Acme Plumbing,” ChatGPT recites your homepage back at you, and you conclude everything’s fine.

That test proves nothing. Nobody who hasn’t heard of you types your name.

Test the questions people ask before they know you exist:

• “Who’s the best [service] company in [your city]?”
• “I need [specific problem] fixed in [your city] — who should I call?”
• “Recommend a reliable [trade] near [neighbourhood]”
• “What should I look for when hiring a [service] in [city]?”
• “[Competitor name] vs alternatives in [city]”

Write down five to ten of these. Use the words your customers use, not the words on your services page. “Blocked drain” not “drainage remediation solutions.”

Step 2: run them in more than one place

ChatGPT is not the whole market, and each assistant is pulling from a different index. Run the same prompts through:

• ChatGPT (with search on)
• Perplexity — heaviest on live citations, so it shows you which pages it trusts
• Google AI Overviews — just search normally and read the AI block
• Gemini and Microsoft Copilot if you want the full picture

Use a logged-out or incognito session. Otherwise your own history quietly biases the answer toward the business you’ve been researching all week — yours.

And run each prompt two or three times. These systems aren’t deterministic; the same question can produce different names on different days. What you’re measuring isn’t a ranking, it’s a frequency. Named in four of five runs is a real position. Named once is noise.

Step 3: score what comes back

There are only three outcomes, and each one means something different.

You’re named. Good — now look at how. Is the description accurate? Does it cite a page you control, or a directory listing, or a review site? If the assistant is describing you through a third-party profile, that profile is doing your marketing for you, and you don’t control it.

You’re not named, but competitors are. The useful question isn’t “why not me” — it’s “what do those three have that I don’t?” Usually it’s boringly consistent: more reviews, clearer service-area pages, a presence on the local directories and roundup posts the model learned from. AI assistants pick brands they can describe confidently, and confidence comes from repetition across sources.

Nobody local is named at all. The assistant hedges and tells the user to “search Google Maps.” That’s the best news on this list: the category is wide open in your area, and being the first business that’s easy to name is a real, cheap advantage.

Step 4: check whether AI can actually read you

If you’re invisible, rule out the dumb causes before the strategic ones.

Your robots.txt. OpenAI runs several separate crawlers, and they do different jobs. OAI-SearchBot is the one that surfaces sites in ChatGPT’s search results; GPTBot is the one that collects training data (OpenAI documents all of them). Plenty of sites blocked “AI bots” in a panic a couple of years ago and quietly took themselves out of ChatGPT’s search answers in the process. Those are two different decisions. Make them separately.

Your key facts in plain text. If your address, service area, phone number and prices only exist inside JavaScript widgets, images, or a PDF, assume they don’t exist. Text on the page, or it didn’t happen.

Your one-line identity. Can a machine extract what you do, where, for whom from your homepage in one sentence? “Family-run plumbing and septic company serving Canton, GA” is quotable. “Excellence, delivered” is not. An llms.txt file is a cheap way to hand that over explicitly.

What not to waste money on

Some honesty, since this is where the industry is currently getting silly.

There is no “rank #1 in ChatGPT.” There’s no admin panel, no submission form, no paid placement that puts you into an organic AI answer. Anyone selling you a guaranteed position is selling you a guess.

Dashboards that track “AI visibility” can be genuinely useful at scale — if you’re a national brand running dozens of prompts weekly. If you’re a local business with one location, a spreadsheet and a monthly half-hour of the test above will tell you the same thing for free.

And the deeply unglamorous truth: the things that get you named by AI are mostly the things that were already working. Consistent business information everywhere, a steady flow of real reviews, pages that answer actual questions in plain language. GEO is less a new channel than a new reason to stop neglecting the old ones.

Do it this quarter

Block thirty minutes. Ten prompts, three assistants, one spreadsheet. Note who gets named and how often. Repeat next quarter and you’ll have something almost nobody in your market has: a trend line for how machines describe you.

If you’d rather not do it yourself, tell us what you sell and where and we’ll run the audit and send you what we find. And if it turns out you’re already the answer in your area, we’ll tell you that too — and tell you to spend the money somewhere else.