How to test whether ChatGPT recommends your business, and how to read the result honestly

Anybody can find out what ChatGPT says about their business in ten minutes. The hard part is not running the test, it is reading the result without fooling yourself, and almost every free checker in this category encourages exactly that.

This page gives the protocol, the baseline from our own measured run, and the four outcomes with what each one actually means.

Run the test properly, in five steps

1. Write the questions a buyer would ask, not questions about you. This is the whole test. Ask what the best option is for a business like your customer, with a constraint in it, a budget, a city, a team size or a requirement. If your company name appears in the question, the answer will contain your company name and you have measured nothing.
2. Keep branded and unbranded questions in separate lists. Branded questions are for checking accuracy. Unbranded questions are the only ones that can measure whether you get recommended. Never average the two into one score.
3. Record the conditions. Date and time, which product, logged in or out, which market, and whether the answer searched the web. For a searched answer, note the cited URLs. In our own run, 550 of 635 readings ran a live search and the rest answered from what the model already held, and the two kinds have completely different source lists.
4. Run each question at least three times. Identical prompts do not return identical answers. One reading is a snapshot. Three readings per question is where it starts to be a measurement.
5. Count it as a fraction. Readings in which you were named, over readings where any company was named at all. Write the date next to it. That sentence is your result, and it is the only form of it worth quoting to anybody.

The denominator almost everyone gets wrong

Many category questions come back as advice with no companies in them. In our own run of 635 readings on ChatGPT and Gemini, 184 named no company at all. For those questions there was no winner, nobody beat you, and nothing you publish would change the outcome, because the question was not a naming question.

So the denominator is eligible readings, meaning the ones where a company was named. Scoring against every reading makes every brand in a category look weak, including the market leaders. Scoring against eligible readings tells you whether you are in the running when the running exists.

The four results, and what each one means

You were recommended

Good, and provisional. Check whether the question had your brand in it, whether the answer searched, and whether it holds across three runs. A single win on a searched answer can come from one newly indexed page and can go away next week.

You were mentioned but not recommended

The most common near miss. It usually means you were retrievable and not corroborated, which is to say you exist in the index and not enough independent sources describe you the same way. This is the gate that content volume does not open.

Competitors were recommended instead

Normal, including for the companies selling the measurement. In our run a rival was named and we were not in 451 of 635 readings. The useful move is to read what the answer cited, because the cited sources are the thing you can actually act on.

No company was named, or the check failed

Score these separately and never as a zero for you. A failed check and a reading that named nobody are two different events, and a tool that silently counts either as an absence will show your visibility falling on a week when nothing happened.

Our own baseline, for calibration

Measured on 22 October 2026. 635 readings across ChatGPT and Gemini, 324 US buyer questions in 19 batches, 550 with live search. 1,053 namings across a roster of 24 vendors, counted once per reading in which a name appeared. 19,793 citation URLs across 3,307 distinct hosts, with half of all citations coming from just 51 of those hosts.

Our own result in that run was 0 namings and 0 citations. Two days earlier, on a different bank of 1,260 readings, it was 0 namings and 6 citations. We publish that because a vendor that will not show you its own worst reading should not be trusted with yours.

What our free scan does and what we do not publish about it

Our pricing page renders a free scan on your own domain, states that the free plan stays free and that every plan starts with the same free scan. What that page does not render is the free scan's exact question limit, its brand count or its engine coverage. We are not going to state a number our own published page does not carry, and you should ask any vendor, including us, for those three figures before treating a free result as representative.

The same caution applies to every free checker in this category. A three-question sample is a three-question sample. It is a result for those three questions on that date, not a probability that an assistant recommends your business.

When a one-off check is enough, and when it is not

A one-off check is enough to answer two questions. Does any company get named for the questions your buyers ask, and is anything the assistant says about you factually wrong. Both are worth knowing and neither needs a subscription.

It is not enough to answer whether anything you did worked. That needs the same questions, held constant, run repeatedly, before and after, with the dates and the run counts published, and ideally a comparison group that changed nothing over the same period. We tested that on ourselves by publishing 507 more pages between 20 and 22 October 2026, going from 134 recommendation-shaped pages to 641. Our naming count went from 0 to 0.

What we charge, and the number that is not the floor

Published US prices are SEO Start 149, Growth 299, Scale 599, SEO Scale 699, Inbound 1099, Outbound 1299, Agency 1299 and Everything 1799 dollars a month. SEO Start reads 0 questions, 0 credits and 0 engines, so AI visibility does not start at 149 dollars, and anybody writing that sentence has read the price column and not the allowance. Growth at 299 is the cheapest plan that reads all five surfaces we cover. A Claude check consumes 3 credits rather than 1.

Our pricing page contradicts itself in three places and both readings belong here. The Agency plan is described as covering 10 client brands while the comparison table answers No to more than one business in all eight columns. The main listing names five apps including Claude while the Scale description names four without it. The FAQ says SEO Start gets 12 pages a month and no AI app checks, beside a table that lists five engines for the same plan.

Where a rival is ahead, by name. We read five surfaces, not nine, and that page renders no SSO, no SAML, no SOC 2, no ISO, no DPA and no white label, checked by word boundary over the whole 119,203-byte document. Read on 22 October 2026, tryprofound.com/pricing publishes up to 9 answer engines tracked plus SSO, SAML and SOC 2 compliance on its Enterprise plan, with no numeric price published, and its free trial covers ChatGPT, Gemini and Google AI Overviews with 50 prompts daily for 7 days.

Questions about testing your business in ChatGPT

Can I just ask ChatGPT what it thinks of my company

You can, and it will tell you about your company, which measures nothing about whether buyers would find you. Use questions that describe the problem instead.

Why did I get a different answer five minutes later

Generation is probabilistic, and the decision to search the web is taken per question, so one run can read live pages and the next can answer from memory. That is why three runs beats one.

Is a visibility score out of 100 worth anything

Only with its formula and its denominator printed. A blended score cannot be checked, and it moves whenever the vendor changes the blend.

Should I test logged in or logged out

Logged out, in the market you sell to. Your own account carries memory, custom instructions and history, which makes it the least representative place to test.