Does ChatGPT know your business? Check in five prompts.

There's one way to find out: ask it. The five prompts below take about fifteen minutes and end in one of exactly three results — ChatGPT describes your business accurately, describes it wrongly, or doesn't know it exists. Each result has a different next move, and all three are covered on this page. One honest warning before you start: AI answers drift from run to run, so a single check is a snapshot, not a verdict. Run the check tonight — then read the last two sections before you celebrate or panic.

Three ground rules before you paste anything

The check only means something if you control what's being tested. Three rules:

1. Use a temporary chat, or log out.ChatGPT's memory, your custom instructions, your location, and your chat history all shade its answers. You've probably typed your own business name into it before — which means your logged-in session already knows you in a way a stranger's doesn't. You want the answer your buyer gets, not the answer your history earns you.
2. Run every prompt twice: web search off, then on. Search off tests the model's trained memory of you — the public web as it existed months ago. Search on tests live retrieval: pages OAI-SearchBot, the crawler OpenAI documents for ChatGPT Search, was allowed to read, surfaced by an index OpenAI does not publicly name. Outside testing is how anyone knows anything about that step — in Seer Interactive's 2026 analysis, roughly 87% of ChatGPT's citations also appeared in Bing's top results, a measured overlap rather than a confirmed pipeline. The two routes fail differently, and knowing which one fails you is half the diagnosis. (The full mechanics are in how ChatGPT recommends businesses.)
3. Record everything, verbatim. Date, prompt, mode, every business named, every source cited. Screenshot the answers. If it turns out ChatGPT is wrong about you, a dated record of the wrong answer is the starting line for fixing it — and the proof later that the fix landed.

The five prompts, ready to paste

Swap your details into the brackets. Where a worked example helps, we use Vista Dental Group in Scottsdale — a fictional business from our example library, not anyone's real scan data. (If you are in that field, the category-specific version of this check lives in AEO for dentists.)

Prompt 1 — the brand-knowledge prompt

“What do you know about [business name] in [city]? What do they offer, and who are they best for?” This is the existence test. A confident, accurate paragraph means the engine knows you. Hedging — “I don't have specific information about…” — means you're absent from its memory. A confident paragraph with wrong services, wrong locations, or another company's history mixed in is the third result, and arguably the most urgent one.

Prompt 2 — the category prompt

“Who are the best [category] in [city]? Give me specific names and what makes each one worth considering.” This is the prompt your buyers actually type, so it's the one that decides revenue. Note every business named and in what order. Being known (prompt 1) but never named here is the common case — the engine can describe you when asked directly, yet never thinks of you unprompted.

Prompt 3 — the service prompt

“I need [a specific service you sell] in [city]. Who should I call?” For the example: “I need Invisalign in Scottsdale. Who should I call?” Category prompts test your brand; service prompts test whether the engines connect you to the specific jobs that pay best. Plenty of businesses show up for their category and vanish for their most profitable service.

Prompt 4 — the comparison prompt

“How does [business name] compare with [your closest competitor] for [service]?” Buyers ask this at the decision stage, and the engine will answer it from whatever public material exists — reviews, roundups, forum threads. If the answer reads like your competitor's brochure, you've learned which sources feed the engine's picture of your market. What to do with that is its own discipline: reverse-engineering why AI picked your competitor.

Prompt 5 — the trust prompt

“Is [business name] legit? What do people say about them?” The nerve-wracking one. Buyers paste exactly this before handing over money, and the engine summarizes whatever reputation material it can find. An empty answer here is almost as costly as a negative one — “I couldn't find much about them” reads as a red flag to a cautious buyer.

The three results, and the next move for each

Result one: it knows you, and gets you right

Genuinely good news — and worth protecting rather than filing away. The answer that names you today is assembled mostly from third-party material: citation studies — including one the monitoring vendor Octolens published from its own dataset — put roughly nine in ten sources cited about a brand on pages the brand doesn't own. Those sources churn — eMarketer's 2026 analysis puts monthly turnover of AI-cited sources between 40 and 60% — so today's clean answer rests on ground that moves. The protective moves: keep your name, category, and facts consistent everywhere they appear, keep your structured data current, and re-check on a schedule rather than once. If the comparison prompt named your rival first even while prompt 1 described you kindly, start with the competitor forensics.

Result two: it knows you, but gets things wrong

Wrong hours, discontinued services, a merger that never happened, or your business blended with a similarly named one two states over. This result is more common than owners expect:

The causes are usually one of three: the training snapshot is months old and your business changed; your identity is entangled with a similarly named company; or so little authoritative material about you exists that the model fills the gaps. One thing that does notwork: arguing with the chat. Correcting ChatGPT inside your session changes your session — nobody else's. The fix happens on the open web, where the engine reads. The step-by-step is in what to do when ChatGPT is wrong about your business.

Result three: it has never heard of you

Hedged non-answers on prompt 1, absence from prompts 2 and 3, and an empty shrug on the trust prompt. Two things worth saying plainly. First, this is not a judgment of your business — engines judge documentation, not quality, and an excellent business nobody has written about is invisible to them. Second, it's structural, which means it's findable and fixable: the causes run from a crawler block on your own site to an empty Bing index to zero third-party corroboration. The four-test diagnostic finds which cause is yours in about ten minutes, and the playbook for getting mentioned by AI orders the fixes by time-to-impact.

Why one clean check can mislead you

Here is the part most “just ask ChatGPT about your brand” articles skip, because it complicates the tidy ending. Three separate mechanisms make a single manual check — pass or fail — a weaker signal than it feels like.

Answer drift. Engines compose each answer fresh, with randomness built in. The same prompt, the same evening, can return a different set of names:

A Tuesday check that finds you and a Thursday check that doesn't can both be “correct” — nothing about your business changed in between.

Personalization.Even in a temporary chat, the engine knows your rough location and tailors accordingly — and any check you run logged in carries your memory, instructions, and history with it. Your answer is not your buyer's answer, and no amount of care fully closes that gap from a single account.

Sampling.Five prompts is a sliver of the ways buyers actually phrase the question — best-of, near-me, cheapest, “for anxious patients”, “open Saturdays” — and one engine is a sliver of the market. Nightwatch's share-of-voice analysis of large prompt sets found the major engines agree on their answers less than half the time, so passing on ChatGPT says little about Gemini, Perplexity, or Google's AI Overviews.

None of this makes tonight's check pointless. It reliably answers the existence question, and it catches wrong facts. It means a pass is provisional and a fail is one data point — and that the difference between checking and measuring is the next section.

What checking at scale actually looks like

The fix for a misleading snapshot isn't a bigger snapshot — it's repetition and breadth. Here is the manual version, honestly, since everything in it can be done free:

1. Build a prompt sheet: 30–50 buyer questions. Not variations of your brand name — the questions buyers ask when they don't know you exist yet. Mix category prompts, service prompts, comparison prompts, trust prompts, and the odd specific one (“open on weekends”, “takes my insurance”).
2. Run the sheet across three or four engines, each prompt in a fresh session, modes noted — ChatGPT, Gemini, Perplexity at minimum, since they disagree with each other often enough that any one of them is a minority report.
3. Score every answer the same way: named or not, cited or not, which competitors appeared, in what order.
4. Compute two numbers.Mention rate: the share of answers that name you. Share of voice: your mentions against each competitor's across the same set. These two numbers, tracked over time, are the actual scoreboard.
5. Repeat on a cadence. Weekly at minimum, daily in a contested market. The repetition is the method — one pass has the same drift problem as one prompt.

Now the honest arithmetic: 40 prompts across three engines is 120 answers to run, read, and score — a serious afternoon, every single week. Most owners run it once, get a number, and stop. Which is a snapshot again, with better paperwork.

And don't expect your analytics to do the watching for you: one marketing firm's analysis — Omni Online Strategies — found only about 20% of ChatGPT brand mentions include a clickable citation link. Single-vendor number, so hold it loosely; the structural point holds regardless, which is that most of what AI says about you never touches a referral report. (What GA4 can catch, and how to configure it, is in how to track AI traffic in GA4.) If you're weighing whether the whole effort is worth making at all, the skeptic's case for and against AEO is the fair place to start.

What AEO Action sells — read this before the CTA

The publisher of this checklist, AEO Action, sells the automated version of the methodology above. The free audit runs the check at machine scale once: ten real buyer questions across ChatGPT, Gemini, and Perplexity — 30 live answers in about two to three minutes, no account, no email, no card — with the three most revealing answers shown word-for-word and every business name highlighted. The PDF is permanent; the shareable link lives seven days. Paid plans run the loop daily — 50 buyer questions per scan on Solo, 100 on Solo Pro, regenerated each scan from your business profile, across your pick of up to six engines (ChatGPT, Gemini, Perplexity, Google AI Overviews, Microsoft Copilot, and Claude — Claude available from Solo Pro up, tracked in model mode and labeled that way, never counted as live). Every scan decomposes into the same numbers you'd compute by hand — mention rate, citation rate, coverage, competitor pressure — with a plain-English work report every seven days. The five prompts above are free forever either way; the methodology is true whether or not you ever automate it.