How to Check if ChatGPT Recommends Your Brand (and Perplexity, Gemini, Claude)
Short answer: Write 10 questions your buyers would actually ask. Ask each one in a fresh, unpersonalized chat in ChatGPT, Perplexity, Gemini and Claude, and repeat it two or three times. For every answer, record whether your brand is named, where it sits in any list, and which competitors appear. Then reduce the results to three numbers: mention rate, average position and competitor share. Repeat the same questions on a fixed schedule. One answer is a sample. The trend is the measurement.
The rest of this guide covers each step, the mistakes that make manual checks misleading, and how to decide whether you need a tool.
Why one screenshot isn't an answer
Most people check once: they type "best [category] tools" into ChatGPT, see whether they're named, and screenshot it. The problem is that AI assistants don't give the same answer twice.
Recomma ran a study in October 2026 that measures this. They asked chatgpt.com the same 100 buying questions twice, within seconds of each other. The two answers named the same brands 78% of the time, picked the same top brand 73% of the time, and cited the same pages only 30% of the time. So if you show up in one answer, it doesn't follow that you show up most of the time. If you're missing from one answer, it doesn't follow that you're invisible.
That's why the method below uses repeated runs and a fixed question set. You want a rate you can compare from one week or month to the next, not a single yes or no.
Step 1: Write the questions your buyers ask (10 is enough to start)
Don't start with your brand name. Asking "What is [your brand]?" tests whether the assistant knows you exist. It doesn't test whether the assistant would recommend you to someone who has never heard of you. Use a mix of four question types:
| Type | Example | What it tells you |
|---|---|---|
| Category | "What are the best invoicing tools for freelancers?" | Whether you're on the default shortlist |
| Use case | "I run a 3-person design studio. What should I use to send invoices and track late payments?" | Whether you're tied to a specific job, not just the category |
| Alternative | "What's a cheaper alternative to [main competitor]?" | Whether you show up in head-to-head buying moments |
| Problem | "How do I stop clients paying invoices late?" | Whether you appear before the buyer knows the category |
Write the questions the way a person types them: with constraints ("for a small team", "under $20", "for Shopify"), not as keyword strings. Then add one or two brand questions ("What is [brand] and who is it for?") to check whether the facts are right. Score those separately, because they don't measure recommendation.
Freeze the list. Once you start tracking, keep the same wording. A change in the questions looks exactly like a change in visibility.
Step 2: Set up a clean test in each assistant
Your own chat history can bias the answer, especially if you've talked to an assistant about your company before. For each assistant:
- ChatGPT: use a temporary chat and choose Unpersonalized before you send the first message. According to OpenAI, an unpersonalized temporary chat "does not use memory, custom instructions, or plugins". You can't change that setting once the chat has started.
- Perplexity, Gemini and Claude: use a logged-out session or a private browser window where possible. If you have to stay logged in, turn off any memory or personalization features first.
- One question per new chat. Earlier questions in a thread shape later answers.
- Note whether web search was used. If an answer shows source links, it drew on live search. If it didn't, it came from what the model learned in training. A new brand often shows up with search but not without it, which is normal and tells you something different from being absent entirely.
- Keep the conditions constant. Use the same day, location and language for a full run, so the results are comparable.
Step 3: Run each question two or three times
Ten questions × four assistants × three runs is 120 answers. That's a lot to do by hand. A practical first baseline is your five most important questions, three runs each, across all four assistants: 60 answers in one sitting. Add the rest later.
If you only have time for one run, label your results "single run" so you don't over-read them later.
Step 4: Record the same fields every time
A spreadsheet works. Use one row per answer:
| Column | What to write |
|---|---|
| Date | When you ran it |
| Assistant | ChatGPT / Perplexity / Gemini / Claude |
| Question | The exact frozen wording |
| Run | 1, 2 or 3 |
| Web search? | Yes / No |
| Mentioned? | Yes if your brand name or your domain appears anywhere in the answer |
| Position | If the answer is a numbered list, the line number you appear on. Leave it blank if you're mentioned in a paragraph without a number |
| Competitors named | Every other brand in the answer |
| Facts wrong? | Old price, wrong category, features you don't have |
| Sources cited | The URLs, if any |
The Sources cited column is the one people skip and later wish they hadn't. Those pages are the shortlist of places where getting covered would change the answer.
Step 5: Turn the rows into three numbers
Mention rate: the answers that mention you divided by all answers, per assistant and overall. Example: you appear in 14 of 60 answers, so your mention rate is 23%.
Average position: the average line number across the answers where you appear in a numbered list. Lower is better. Report how many answers it's based on ("avg. #3.2 across 9 ranked answers"), because a position based on two answers means little.
Competitor share: for each competitor, the share of answers that name them. This is your real competitive set in AI answers, and it often isn't the set you'd draw on a whiteboard. If a brand you've never worried about appears in half the answers, that's a finding in itself.
Example of the summary you'd end up with (illustrative numbers, not real data):
| ChatGPT | Perplexity | Gemini | Claude | All | |
|---|---|---|---|---|---|
| Mention rate | 4/15 | 2/15 | 5/15 | 3/15 | 14/60 (23%) |
| Avg. position (ranked answers) | #3 (n=3) | n/a (n=0) | #4 (n=4) | #2 (n=2) | #3.2 (n=9) |
| Top competitor named | Brand A 12/15 | Brand A 13/15 | Brand B 10/15 | Brand A 9/15 | Brand A 70% |
Step 6: Read the result and decide what to do
Most brands land in one of four situations:
- Not named in category or problem questions at all. The pages assistants draw on don't connect you with your category. Look at the sources column. The fix usually lives on other people's sites: comparison articles, directories, review sites, community threads. It rarely lives on your own homepage.
- Named only when the question includes your name. The assistant knows you but doesn't consider you a default choice. Use-case and alternative questions are often the easier wins here.
- Named, but with wrong facts. This can be worse than not being named. Find where the wrong fact comes from (an old review, an outdated listing) and correct it at the source. Make sure your own site states your current pricing and positioning plainly.
- Named, but a competitor is always first. Read the cited pages behind the answers where they rank first. Often a small number of list articles drive the whole answer.
Change one thing at a time where you can, then re-run the same questions. If you change five things at once and the mention rate moves, you won't know why.
Step 7: Repeat on a schedule you'll actually keep
Monthly is enough for most small brands that aren't actively changing anything. If you're publishing, pitching or fixing listings, a weekly check tells you sooner whether the work is landing. Either way, compare the same questions under the same conditions, and judge movement over several runs, not one.
Common mistakes
- Testing only ChatGPT. Assistants differ a lot. You can be named often in one and never in another. Checking only one tells you about one.
- Asking with your brand in the question. "Is [brand] good for X?" almost always names you. It measures recall, not recommendation.
- Rewording questions between runs. You'll measure your own edits.
- Treating position in a paragraph as a rank. Only numbered lists have a meaningful order.
- Ignoring wrong facts because you were "mentioned". A mention with the wrong price can still lose you the deal.
Do you need a tool for this?
Honestly, often not at first. If you have one brand, a handful of questions and the discipline to re-run them monthly, the spreadsheet above is enough. It's also the best way to learn which questions matter before you pay to track them.
A tool earns its keep when the manual version breaks: you want a weekly cadence, you're tracking more than about 10 questions, or you're an agency doing this for several clients. Published prices vary a lot. Otterly.AI's Lite plan is $29/month for 15 prompts tracked daily across ChatGPT, Google AI Overviews, Perplexity and Copilot, with Claude and Gemini as paid add-ons. AthenaHQ's Starter plan is $295/month for 3,600 AI responses across 11 models. Profound offers a free 7-day trial and custom Enterprise pricing. (Prices checked October 27, 2026.)
Recallio, the tool we're building, automates the core of this method (fixed questions, all four assistants, mention rate, rank and competitors, on a weekly schedule). It runs each question once per assistant per week and doesn't collect cited sources, so keep a manual sources check if you need it. Each week it asks ChatGPT, Perplexity, Gemini and Claude your questions, then sends a fresh report about every 7 days (and you can trigger one on demand) with your mention rate, rank, which of your competitors they named, and what changed since last week. Solo is $19/month for one brand and up to 20 prompts. Agency is $49/month for five brands. Two trade-offs to know about: it runs weekly, not daily, and it uses the official APIs, not the consumer chat apps. API answers aren't identical to what a person sees in the app. The same Recomma study found the API's brand overlap with chatgpt.com was 65%, against 78% for the interface compared with itself. That makes API tracking useful for trends but not a perfect copy of the app experience. Recallio is pre-launch, so if that fits, you can join the waitlist.
FAQ
How do I check if ChatGPT recommends my business?
Open a temporary, unpersonalized ChatGPT chat and ask the questions your buyers would ask, such as "best [category] for [use case]" or "alternatives to [competitor]", without naming your business. Ask each question two or three times in new chats and record whether you're named, your position in any numbered list, and which competitors appear.
Why does ChatGPT give a different answer every time I ask?
AI assistants generate each answer fresh, and when web search is used the sources can change between runs. In Recomma's October 2026 test, two identical chatgpt.com questions asked seconds apart named the same brands 78% of the time and cited the same pages only 30% of the time. That's why a single answer should be treated as a sample, not a measurement.
What is an AI mention rate?
AI mention rate is the share of AI answers that name your brand or your domain out of all the answers you checked. For example, appearing in 14 of 60 answers is a 23% mention rate. Calculate it per assistant and overall, using the same questions each time.
How many questions should I track?
Start with 5 to 10 questions that mix category, use-case, alternative and problem questions. That's enough to see patterns. Add more only once you know which ones matter, and keep the wording fixed so results stay comparable.
Should I check Perplexity, Gemini and Claude too, or just ChatGPT?
Check all four if your buyers might use them. Assistants draw on different sources and often name different brands, so visibility in one doesn't predict visibility in another.
How often should I check?
Monthly is enough for most small brands that aren't actively changing anything. Weekly makes sense when you're publishing, pitching or fixing listings and want to see sooner whether it's working. Judge changes over several runs rather than one.
Can I pay ChatGPT to recommend my brand?
Not in the answer itself. ChatGPT now shows labeled ads on some plans, but OpenAI says ads "do not influence ChatGPT's answers" and advertisers "have no ability to shape, rank, or alter" responses. There's also no submission form for organic recommendations. What you can influence is what assistants find about you: accurate facts on your own site, and coverage on the third-party pages they cite.