Is My Business Showing Up in AI Search? (How to Check)
To check whether your business shows up in AI search, test the three major assistants — ChatGPT, Gemini and Grok — against the real buyer questions your customers would ask, not just your brand name. Ask each model neutral, category-level prompts (“best [category] for [use case] in [place]”) in a fresh session, and note whether you’re named, ranked, or absent, and which competitors appear. Don’t test only your own name — every business shows up when you ask about it directly; what matters is whether you surface when the buyer doesn’t already know you. Manual testing is useful but slow and easy to bias. A faster, less biased route is a free Domain Check, which reads a continuously updated index and returns the actual query list your domain is cited on across all three models.
The manual method (and why it’s harder than it looks)
The do-it-yourself approach is straightforward in theory: open ChatGPT, Gemini and Grok, ask the questions a buyer would ask, and see if you’re named. In practice, a few traps make casual testing misleading. The method below avoids them.
1. Test buyer questions, not your brand name
Asking “what is [your business]?” almost always returns something — the model is summarizing your own site. That tells you nothing about discovery. The real test is whether you appear when someone describes a need: “best project management tool for small agencies,” “reliable plumber in Leeds,” “affordable CRM for solo consultants.”
2. Use fresh, signed-out sessions
Assistants personalize. If you’ve been chatting about your own company, memory and context can nudge it to mention you. Use a new chat, and where possible a signed-out or temporary session, so you see what a stranger sees.
3. Vary the phrasing
Buyers ask the same thing many ways. Test several phrasings of each need — with and without your city, with and without a price qualifier — because a model can cite you on one phrasing and ignore a near-identical one. This variance is normal and is part of why a single test can be misleading.
4. Record what you see
For each query and each model, note: were you named? Ranked high, low, or only if prompted? Which competitors appeared? Over a dozen queries a pattern emerges — the questions you reliably win, the ones you’re absent from, and the rivals the models treat as your substitutes.
The self-test prompt table
Don’t improvise. Run the same structured set against each model so your results are comparable. Replace the bracketed parts with your own category, audience, place and brand. For each row, note whether you were named, ranked, or absent — and who appeared instead.
| Test query | Model | What to look for |
|---|---|---|
| Best [category] for [audience] in [city] | All three | Are you named in the shortlist, and which competitors appear? |
| Affordable / reliable [category] for [use case] | All three | Do you survive a price/quality qualifier, or only generic phrasing? |
| [Service] near me | All three | Does the model ask for or assume a location, then who does it name? |
| Alternatives to [a known competitor] | All three | Are you listed as a credible alternative, or omitted entirely? |
| Who should I hire for [specific problem] in [city]? | All three | High-intent phrasing — being named here matters most. |
| What is [your exact brand name]? | All three | Control row only: almost everyone passes this — it proves nothing about discovery. |
What the results actually mean
Don’t over-read a single absence. The models disagree with each other and even with themselves between sessions — see ChatGPT vs Gemini vs Grok: how each picks businesses. What matters is the pattern across many queries and all three models: consistent presence is a real asset, consistent absence on high-intent questions is a real gap, and disagreement between models is a clue about which signals each one trusts.
Why manual checking hits a wall
Manual testing is honest work, but it’s slow, it samples only the questions you thought to ask, and it’s hard to keep neutral. You can’t feasibly test hundreds of phrasings across three models by hand every month, and you’ll naturally test the queries where you expect to do well. That sampling bias is exactly what hides your real gaps.
Copy-paste self-test prompts
Paste these straight into ChatGPT, Gemini and Grok (one per fresh chat), swapping the bracketed parts for your details. They are written to surface discovery, not flattery:
| Copy-paste prompt | Why this prompt |
|---|---|
| I'm looking for the best [category] for [audience] in [city]. Give me a shortlist with a sentence on each. | The core discovery test — are you in the shortlist at all? |
| Recommend a [category] for someone who needs [specific use case]. Name three options. | Forces a concrete shortlist tied to a real need. |
| What are good alternatives to [competitor] for [use case]? | Checks whether the model treats you as a peer of a known rival. |
| I need a [service] near me in [city] this week — who would you suggest? | Tests local + urgency resolution the way buyers actually ask. |
| Compare [your brand] and [competitor] for [audience]. Which would you pick and why? | Reveals how the model frames you against a direct competitor. |
Manual prompting is honest but slow and easy to bias toward queries you expect to win. For a complete, unbiased list, use the query-level check below.
The faster, less biased route: a query-level check
Instead of sampling a few prompts, MentionRadar runs buyer questions through all three models continuously and records which domains get cited or mentioned in each answer. The free Domain Check reads that index backwards from your domain and returns the real list of queries you appear in — with the intent behind each, which models named you, and the competitors named alongside. You get the worklist a single visibility score never could.
The mechanic behind it — the inverted query–domain index — is explained in the Reverse AI Search pillar. Once you know where you stand, move on to how to get your business recommended by ChatGPT or return to the small-business pillar. For the theory behind why answer engines surface some sources over others, see the AEO/GEO fundamentals pillar.
Frequently asked questions
What should I type to test if I show up in ChatGPT?
Test category-level buyer questions, not your brand name — for example “best [category] for [use case] in [city]” — in a fresh, signed-out session, and note whether you are named and which competitors appear.
Why does ChatGPT mention my business when I ask about it directly but not otherwise?
Asking “what is [your business]?” just makes the model summarize your own site, which proves nothing about discovery. The real test is whether you surface when a buyer describes a need without knowing your name.
Why do I get different results each time I ask?
Assistants personalize and vary between sessions, and the three models weight different signals — see how each model picks businesses. Judge the pattern across many queries, not one answer.
Is there a faster way than testing prompts by hand?
Yes. A free Domain Check reads a continuously updated index and returns the real query list your domain is cited on across all three models, without the sampling bias of manual testing.