The short answer
Ask each engine the questions your buyers ask, not your brand name. Note whether you're mentioned, where you rank and which sources it cites. Then rephrase and ask again, because one answer tells you very little.
Search for your own brand name in ChatGPT and you’ll almost always get something back. That isn’t the test that matters. What matters is whether you show up when a buyer asks for a recommendation and doesn’t know your name yet.
Step 1: Write 10 prompts your buyers would type
Mix four kinds of question. Write them the way a customer talks, not the way your homepage does.
- Category
- What are the best [category] tools for [audience]?
- Comparison
- What are good alternatives to [competitor]?
- Problem
- How do I [job your product does] without [pain]?
- Brand facts
- What is [your brand] and who is it for?
Aim for three category prompts, three comparisons, three problems and one brand-facts prompt. Swap the brackets for your own market.
Step 2: Run them in each engine separately
Each assistant draws on different sources and answers in its own way, so what one says tells you little about the others. Use a logged-out or fresh session where you can, so your own history doesn’t steer the answer.
| Engine | What to watch for |
|---|---|
| ChatGPT | Whether it searched the web for this answer or answered from what it already knows. The two can disagree. |
| Perplexity | The numbered sources it cites. They show which pages it trusts for your category. |
| Gemini | Whether your own site or third-party pages get cited. |
| Claude | Whether web search was on. Log answers with and without it separately. |
Step 3: Record three things per answer
- Mentioned or not. A brand that only appears in a footnote counts as a weak yes.
- Position. First of five is very different from fifth of five.
- Sources cited. Note which sites the answer relies on. Those are the places worth earning a mention.
A simple sheet works: one row per prompt, one group of columns per engine.
Step 4: Rephrase and ask again
Ask the same question three slightly different ways. If you appear in one version and disappear in the next, your visibility is fragile. That’s worth knowing before you change anything.
Why doing this every week gets old fast
One check is a snapshot. AI answers shift as engines update and as new pages are published and cited. To know whether your work is paying off, you need the same prompts on the same engines, on a schedule. By hand, that’s 40 answers to read and log every week.
That’s when tracking it automatically starts to make sense.
Questions
Does ChatGPT give everyone the same answer?
No. Phrasing, conversation history, memory settings, location and whether it searches the web all change the answer. That's why you test several phrasings.
Should I just ask "Do you know [my brand]?"
Ask it once to check that what the engine says about you is correct. It won't tell you whether you'd be recommended to someone who has never heard of you.