Most business owners, when they first suspect they might be invisible to AI, do the obvious thing. They type their own name into ChatGPT and see what comes back.
It almost always gives back a result that feels reassuring and isn’t. Of course the engine can say something about you when you hand it your name. The question is whether it would name you if a potential client typed “who should I hire to do what you do” without already knowing you exist.
That’s the test worth running. It’s a different test entirely, and most people never run it.
How do I check if AI recommends my business?
To check if AI recommends your business, run the question a potential client would ask, not your own name. Open ChatGPT, Perplexity, Google AI Mode (or Google AI Overviews), and Grok in fresh, logged-out windows so you see the cold result a stranger sees, not one shaped by your history. In each, type the question a potential client would actually use: “who should I hire for [your service] in [your location]?”, “best [your type of business] for [your ideal client]?”, and “I need help with [the problem you solve], who do you recommend?” Then read for three things: whether you appear at all, how accurately you’re described, and who is named instead of you. Run the same prompts across at least two engines, because they pull from different sources and weight them differently. Screenshot the results so you can compare them honestly rather than rationalise a bad one from memory.
What are you actually testing when you check AI?
You’re not testing whether AI knows you exist. You’re testing whether AI would recommend you to someone who doesn’t.
The phrasing matters more than most people expect. “Sally Kuter consultant” returns information. “Who should I hire to help me get found by AI” returns a recommendation. The second one decides whether you make the shortlist, and those two queries can come back completely different.
You’re also testing across more than one engine. ChatGPT, Perplexity, Claude, and Google AI Overviews pull from different sources and weight them differently. A good result in one and a missing result in another tells you something different from what a consistent gap across all four tells you.
Plan for about 20 minutes. Do it properly once rather than quickly twice.
Which prompts should you run to check AI recommendations?
Adapt the language to your actual service, location, and ideal client.
Category recommendation prompts (the most important ones)
These are what potential clients type when they’re looking for someone like you.
- “Who should I hire for [your main service] in [your city or country]?”
- “Best [consultant / strategist / coach / specialist] for [the core problem you solve]?”
- “I need help with [your core client problem]. Who do you recommend?”
- “What kind of business helps founders with [the outcome you deliver]?”
Run these word for word first. Then vary them. If you’re a business operations consultant in Brisbane, you want to know what comes back for “operations consultant Brisbane”, “business systems for service businesses”, “help with operational complexity for a founder”, and half a dozen variations in between. Potential clients don’t all phrase the same question the same way.
Problem-led prompts (the ones that reveal who’s in your category)
These reveal which names the engine associates with the category of problems you solve.
- “Why isn’t my business showing up in AI recommendations?”
- “How do I get my business found by ChatGPT?”
- “What is citeability and who helps with it?”
- “How do I improve my AI search visibility?”
If you’re positioning in this space, your name should start appearing in responses to these. If it doesn’t, the content work isn’t landing yet.
Validation prompts (what potential clients run after getting a referral)
These are what a referred client types to check whether you’re credible before they contact you.
- “[Your name]: is she legit? What is she known for?”
- “What does [your name] specialise in?”
- “Is [your name] the right person to help with [the specific thing]?”
These tend to return more generous results because the engine has your name to anchor to. Don’t mistake a good validation result for evidence that the category recommendation prompts are working. They’re measuring something different.
How do you run the prompts across ChatGPT, Perplexity, Claude, and Google?
Open each engine in a fresh window and do not log in. Use private or incognito browsing if you can. Logged-in sessions are personalised. They surface results shaped by your history, which isn’t what a stranger sees. You want the cold result, not the one shaped by what the engine already knows about you.
Run the same prompts across ChatGPT, Perplexity (search mode, not Focus), Google AI Mode and AI Overviews (search for your category query and see if the AI answer appears at the top), and Grok. Add Claude if you want a fifth read. Each pulls from different sources and weights them differently, so the pattern across them is more useful than any single result. When I checked Andre Obradovic’s results, ChatGPT, Perplexity, Google AI and Grok all returned the same verdict at first: silence. That agreement across engines told me it was a signal problem on his side, not a quirk of one platform.
Copy the results. Screenshot them, paste them into a document, or take notes. You want to compare them honestly rather than rely on memory.
Run each prompt twice if the result surprises you. These engines have some variability, particularly for queries where there’s no single dominant answer.
What should you look for in the results?
Are you there at all? If your name doesn’t appear in response to the category recommendation prompts, the engine doesn’t have enough clear, consistent, quotable signal to include you. That’s the first thing to fix.
How are you described? If you appear but the description is inaccurate, outdated, or so generic it could be anyone, that’s a signal problem, not a visibility problem. The engine is working with material that doesn’t reflect what you actually do now. This is common after a pivot or a repositioning.
Who is being recommended instead? Note the names. Usually it isn’t that those businesses are better known. It’s that they’ve written about the specific problem more clearly, or their language lines up more directly with the query, or they show up consistently across more sources. The engine isn’t ranking popularity. It’s ranking legibility.
What reasons does the engine give? If it recommends a competitor, the stated reason often tells you exactly what it read that you didn’t give it. “Jane is recommended for her clear framework on X” means the engine found a clear framework on X. If you have a clear framework and it’s not in the answer, the problem is that it’s not clearly structured or publicly visible.
What’s the trap when you read your own result?
Reading your own result, knowing how good your work actually is, puts a filter over it that doesn’t help you. “Well, I do offer that.” “The engine got that slightly wrong.” “That’s not quite how I’d describe it, but it’s close.” Every one of those is you being generous to a result that a stranger wouldn’t be generous to.
Read it as a potential client who doesn’t know you. Does the result give them a reason to call you? Does it make you sound like the specific, credible expert for their problem? Or does it give them a vague, slightly beige version of an expert that could be anyone in your category?
If it’s the latter, that’s your gap. And now you know it.
A bad first result is not a verdict on your work. It’s a starting number you can move. When Andre Obradovic, a men’s health coach, first ran this exact check, no engine named him on his core client queries. A clean zero, across the board. That’s what a low starting score looks like. After the citeability work, ChatGPT, Perplexity, Google AI and Grok now name him first on most of those same queries. Empty, then named: both ends of the scale on the same person. The full case study is here. A low starting score is the most common result, and it’s the most fixable one.
The next step is understanding why the result looks the way it does. How AI engines decide who to recommend covers what the engine is actually measuring and how to shift the signals in your favour.
If you ran the check and came up empty, or beige, you don’t need to guess at why. The Citeability Snapshot runs the same check properly across the queries that matter for your category, then tells you which signal is missing and what to fix first. You bring the low score. It tells you what’s holding it down.