You check whether AI recommends your business by asking the four tools your customers actually use (ChatGPT, Gemini, Perplexity, and Claude) the same questions those customers ask, once with your business name and once without it. Do it in a fresh or incognito session so past chats don’t skew the result, and log three things each time: whether your name came up, where in the answer, and who got named instead. You can run the whole thing in one evening, for free, before you spend anything on an audit.
The point of testing with and without your name is to separate two states that feel identical from the inside. If AI answers well when you ask about your company directly but goes quiet when you ask a general question, it knows you and doesn’t recommend you. That’s the exact pattern we cover in why ChatGPT doesn’t recommend your business, and it’s the thing this test is built to expose.
How do I check if AI recommends my business?
Ask each tool a small set of questions, twice: a branded version (“what do you know about [your company]”) and a no-name version (“recommend a good [your service] in [your city]”). Run each in a clean session, screenshot the answers, and note whether you appear. The no-name questions are the ones that reveal whether AI recommends you or merely recognizes you.
The method is deliberately low-tech. You don’t need software or a subscription for a first read. You need a consistent set of prompts, clean sessions so the tools don’t lean on your history, and somewhere to write results down. The discipline is what makes the test trustworthy, not the tooling.
Which questions should I ask?
Use five kinds of prompts, because each checks a different slice of visibility. A branded prompt tells you if AI knows you at all. Generic, local, price, and comparison prompts tell you whether you make it into recommendations when a stranger asks. Skip a type and you get a flattering, incomplete picture.
| Prompt type | Example (swap in your own words) | What it reveals |
|---|---|---|
| Branded | “What do you know about [company]?” | Whether AI recognizes you at all |
| Service (generic) | “Can you recommend a good [service] provider?” | Whether you make it into a no-name recommendation |
| Local | “Who does [service] in [city]?” | Visibility where your customers actually are |
| Price | “How much does [service] cost in [city]?” | Whether AI cites your prices and your page |
| Comparison | “Compare the best [service] companies in [city]” | Who you’re ranked against, and whether you’re ranked at all |
Fill the brackets with the words a customer would use, not industry jargon. A dentist’s patient doesn’t ask about “implantology”; they ask “how much to get a tooth implant.” Match the phrasing to the person, or you’ll test a question nobody types.
Why test all four tools, not just ChatGPT?
Because your customers don’t all use the same one, and the four tools reach for different information. A business that shows up strongly in ChatGPT can be absent in Perplexity and Gemini, or the reverse. Testing only your own favorite tool measures your habits, not your market’s. Four reads give you the real spread.
Each has its own tendencies. ChatGPT and Perplexity lean heavily on live web results, and Perplexity shows its sources inline, which is useful for seeing exactly what it read about you. Gemini draws on Google’s understanding of your business. Claude answers from the material it’s given and its training. You don’t need to memorize the differences. You just need to check all four, because your prospects do.
How to run the test step by step
Run the same set the same way every time. Consistency is what turns a handful of screenshots into evidence you can compare next month. The whole loop takes an evening for one tool and an evening or two for all four.
- Open each tool in a private or incognito window, signed out. Signed in, the tools remember your past chats about your own company and will mention you more often than a stranger would see. That inflates your score.
- Ask each prompt in a separate conversation. Questions asked in one thread contaminate each other. One prompt, one fresh chat.
- Log three things per answer: did your name appear, where in the list, and who was named instead of or alongside you.
- Screenshot everything. You won’t reproduce today’s exact answer next month, and you’ll want the before-and-after.
- Repeat the full set after a few days, ideally twice. AI answers shift between sessions, so a single run can mislead. A pattern across runs is the real signal.
- Put it all in one table. Without it, a week later you’re left with a vague “I think it showed me once.”
A working results table can be as simple as this:
| Prompt | Tool | Date | Named? | Position | Named instead |
|---|---|---|---|---|---|
| “Who does [service] in [city]?” | ChatGPT | Jul 17 | No | — | 5 competitors |
| “How much does [service] cost?” | Perplexity | Jul 17 | No | — | 2 directories |
| “What do you know about [company]?” | Gemini | Jul 17 | Yes | 1 | Just us |
How do I read the results?
Read the branded rows and the no-name rows as two separate stories. If AI answers richly about you by name but drops you from every general question, you’ve confirmed the “knows you, doesn’t recommend you” gap. That split is the norm, not the exception, and it points straight at your missing recommendation signals rather than a missing website.
We watched this play out in our own pilot audit. Asked about the business by name, the AI described it correctly. Asked the general, no-name questions a real prospect asks, it left that business out of 8 out of 10 answers. The most telling case was price: a service cost was published right there on the page, and AI still didn’t cite it. The information existed; the citable, trustworthy form did not.
If your no-name rows are mostly empty while your branded rows look great, that’s your finding. The next question is what specifically on your site is blocking the recommendation, and in what order to fix it.
When should I hand this to a pro?
Hand it off when you’ve confirmed a problem and need to know its causes, its size, and what to fix first. The DIY test is excellent at showing symptoms. It can’t see inside what the model reads on your site, it’s hard to reproduce under identical conditions, and covering four tools thoroughly, several times, gets tedious fast. For “do I have a problem,” your evening test is enough. For “which fix pays off first, and did it work,” you need a repeatable protocol.
That’s the line where an AI visibility audit earns its cost. We run the same set of questions across ChatGPT, Gemini, Perplexity, and Claude, log the answers and sources, and score 39 criteria across six areas, then hand you a prioritized fix list with a score you can benchmark against later. It starts at PLN 499 net (around $125), with a refund if we don’t surface at least five things worth fixing. If you’d rather talk it through first, bring your results table to a free 20-minute consultation and we’ll read it together. Want a wider marketing view before you narrow to AI? Start with our free 3-minute marketing audit.
FAQ
Is the free version of ChatGPT enough for this test? For a first read, yes. The free version answers all five prompt types, so you’ll see whether you appear in no-name questions. If you have access to web-search mode, test it separately: the browsing model tends to cite pages directly, while the non-browsing one leans on training data. Those are two different results, and both are worth knowing.
How often should I re-run the check? Run the full prompt set once a month, always under the same conditions: incognito, fresh chats, identical questions. AI visibility is a trend, not a fixed state, so one measurement says little and a series shows direction. After making bigger changes to your site, wait a few weeks before crediting a shift to your work rather than normal variation.
AI shows me once and not the next time. Which result is true? Both. Language models don’t return identical answers to the same prompt, which is exactly why the method builds in repeats. If you appear in one of three runs, your visibility is shaky: the model considers you but isn’t convinced. The goal is to appear every time. How to measure AI visibility covers how to read that variation instead of chasing it.
Can I improve visibility without paying for an audit? Yes, and many businesses should start on their own: let the AI crawlers in, make your business facts consistent, and rewrite key pages so they answer real questions in plain text. An audit becomes worth it when you want to know which of those changes will move the needle most for you, and whether the needle actually moved after a month. For the local angle specifically, see AI visibility for small and local businesses.