How to get ChatGPT to recommend your business: check first, then fix the right thing

When a buyer asks ChatGPT, Perplexity, or Google AI who to hire, the engine answers with two or three names. Ranking in ChatGPT means being one of them. Here is a 20-minute AI visibility check to find out whether you are, and what the answer tells you to fix when you aren't.

By Sean Garschi, Founder, Growthr · Published September 1, 2026

The check, step by step

You need a prompt, a clean session, and a way to record the result. Skip any of the three and you will misread what you see.

1. Write the prompt a buyer would type, not the one you would

Buyers describe a problem. They rarely name the category, and they never name you. A real prompt from a buyer who later hired us read: "I'm looking to hire someone to fix my website for SEO and AI visibility. Please suggest companies and provide pricing and deliverables." Write three to five prompts in that shape for your own business: the pain, the constraint (budget, location, timing), and the ask. Keep the wording exactly the same every time you run the check, or you cannot compare results month to month.

2. Use a session that does not know you

This is the step most people skip, and it produces false positives. If you are signed in to ChatGPT and have ever discussed your company there, its memory will shape the answer around you. Open a temporary chat instead (the toggle at the top of a new chat), which starts with no memory and saves nothing. On Perplexity, run the prompt logged out. On Google, use a normal search and read only the AI Overview at the top, if one appears.

We made this mistake ourselves. A check run from a signed-in account came back with the line "For Growthr specifically, I'd contact..." followed by five competitors. The engine had recognised who was asking and answered a different question. The same prompt in a temporary chat did not mention us at all.

3. Wait for the whole answer

Engines with web search take a while. ChatGPT can take 90 seconds or more on a shortlist question; Perplexity streams and looks finished before it is. Read the answer only once the sources panel has settled.

4. Record the rung, not a yes or no

"Do we show up" has five answers, and they mean different things:

• Absent. Your name is nowhere in the answer or the sources.
• Cited. One of your pages appears in the sources panel. The engine read you but did not necessarily name you.
• Mentioned. Your name is in the text, often in a paragraph about someone else.
• Recommended. You are on the shortlist the buyer is given.
• Recommended against. The answer steers the buyer away from you. Rare, and the most important result you can get.

For each prompt and each engine, write down the rung, who else was recommended, and which domains the sources came from. That last column is the useful one.

What the sources column tells you

An AI answer is assembled from the pages the engine retrieved for that prompt. Look at where they came from and you can see what actually decides the shortlist in your category.

When we ran "which companies can do SEO and GEO for me" in September 2026, ChatGPT built its answer from seven pages: four were the recommended agencies' own service pages, and three were their Clutch review profiles. It ranked them partly by review count, and it stated outright that it would reject any agency whose plan was mostly schema markup and llms.txt. No listicles, no press. In that category, a service page that describes the discipline properly plus a review profile with a handful of verified reviews is what decides the shortlist.

When we ran buyer prompts for a New York event photographer, the sources were different again: venue and vendor directories, review platforms, and a few competitor sites with a lot of text on them. Her own site was clean and ranked on Google, and she appeared in none of the fourteen businesses the engines named, because the directories the answers were built from did not list her.

The sources column is what tells the two cases apart, and it is what sets the fix.

The four reasons you are absent

Almost every "absent" result traces to one of these, and they are worth diagnosing in this order because the fixes get more expensive as you go down.

The engine cannot read you

AI crawlers are blocked by a firewall rule, a bot-management setting, or a robots.txt line someone added in 2024. Or the page loads its text with JavaScript that the crawler never runs, so it fetches an empty shell. This is the cheapest problem to have. You can test it in twenty seconds with a free scan at willaiseeme.com, which fetches your site the way each AI crawler does and shows what came back.

The engine reads you but cannot quote you

Your pages exist and load, but there is nothing on them an answer engine can lift: the services are in an image, the pricing is "contact us", the about page is two sentences, and the FAQ is a set of collapsed accordions with no text in the HTML. Engines cite pages they can extract a specific, self-contained answer from. A page that states what you do, for whom, at what price, in plain sentences, gets quoted. A page that hints gets skipped. This is also where a public price matters: the buyer prompt above asked for pricing, and the answer's table had a pricing column. Companies without a number on their site were listed as "custom quote", and the buyer told us afterwards that the price on our site was why she called.

Nobody else vouches for you

This is the one that keeps sites cited but never recommended. Recommendation is a consensus judgment: the engine names the businesses that show up across the review sites, directories, comparison articles, and industry lists it retrieved. Your own site cannot supply that corroboration, however good it is. If the sources column for your category is full of one review platform or one directory, that is your to-do list, and it is off-site work: reviews, listings, a mention in the comparison article everyone cites.

The engine has the wrong idea of who you are

Less common, but it happens to companies with a generic name or a second line of business. In our own August check, one engine decided Growthr was a product-analytics tool and recommended Amplitude and Mixpanel as alternatives to us. The fix is entity work: the same name, description, and links on every profile you control, organization schema that says who you are, and enough third-party pages that agree.

What to fix first

1. Run the check and keep the log. Five prompts, three engines, the five rungs, the sources column. Twenty minutes, monthly. Everything else depends on this baseline.
2. Unblock the crawlers and put text on the page. The two cheap fixes. A day of work for most sites, and they move the "cited" rung within weeks.
3. Publish your price and your scope. Buyers ask for it, engines tabulate it, and a number on the page is what turned an AI answer into a call for us.
4. Go where the sources are. Whatever platform keeps appearing in the sources column for your category, get listed there and get real reviews there. This is slow and it is the only thing that moves the "recommended" rung.
5. Re-run the check next month. One month's change is noise. Two months in the same direction is signal. Do not rewrite your site on the strength of one bad answer.

Frequently asked questions

How do I check if ChatGPT recommends my business?

Open a temporary chat in ChatGPT so your account memory cannot bias the answer, type the question a buyer would actually ask (a problem, not your brand name), and look for your name in the answer and your domain in the sources panel. Repeat the same prompt in Perplexity while logged out and in Google to see whether an AI Overview names you. Record where you land: absent, cited as a source, mentioned in the text, recommended on the shortlist, or recommended against.

Why does ChatGPT not recommend my business even though I rank on Google?

Ranking and being recommended are different systems. An AI answer is built from the sources the engine retrieves for that prompt, and for most buying questions those are third-party pages: review sites, directories, industry lists, comparison articles. If none of those corroborate you, the engine has no basis to recommend you, whatever your own site says. Being fetched but not cited is also common when a page has no extractable text the engine can quote.

What is the difference between being cited and being recommended by AI?

Cited means one of your pages appears as a source; the engine used your content to build the answer. Recommended means you are on the shortlist the engine gives the buyer. Citation depends on your content being retrievable and quotable. Recommendation depends on consensus across the web about you. Sites often earn citations on questions that already contain their name and still get zero recommendations on questions where the buyer does not know them.

How often should I check my AI visibility?

Monthly, with the same prompt set every time. Answers vary from run to run, so a single result is noise; a two-month trend is signal. Log the rung for each prompt (absent, cited, mentioned, recommended, recommended against), who else was named, and which sources the answer used.

Related: Does llms.txt actually work? · SEO + GEO services · Pricing