How to check what ChatGPT and other AIs say about your brand
To check your AI visibility, ask each major assistant the questions your buyers ask, in a fresh session with web search on, and record whether you are mentioned, how you are described, who else is named and which sources are cited. Repeat monthly with the same prompts so you can see change.
Updated · Citable
Step 1: write your prompt set
Aim for 15 to 30 prompts, written the way buyers actually ask, with their constraints:
| Type | Examples | What it tests |
|---|---|---|
| Brand | "What is [brand]?" "Is [brand] legit?" "How much does [brand] cost?" | Whether the assistant knows you and has your facts right |
| Category | "Best press release service for a startup" "Cheapest way to get a press release on AP News" | Whether you are on the shortlist |
| Comparison | "[brand] vs [competitor]" "Alternatives to [competitor]" | How you are positioned against others |
| Problem | "How do I get my SaaS mentioned by ChatGPT?" | Whether you appear for the problem, before the buyer knows the category |
Step 2: run it cleanly
- Use a fresh chat for each prompt, logged out or in a temporary chat, so memory and past conversations do not influence the answer.
- Turn web search on where it is optional (ChatGPT, Claude), and also try without it for brand prompts: that shows what the model knows from training.
- Cover ChatGPT, Perplexity, Gemini, Claude, Microsoft Copilot and Google (AI Overviews and AI Mode).
- Run your top five prompts two or three times. If you appear in one run out of three, that is useful information.
Step 3: record the same fields every time
| Field | Why |
|---|---|
| Mentioned? (yes/no) | The headline number: share of prompts where you appear |
| Position in the list | First-named brands get more attention |
| Description used | Shows what the assistant thinks you are for |
| Wrong facts | Old prices, wrong category, features you do not have |
| Competitors named | Your real competitive set in AI answers |
| Cited URLs | The sources deciding the answer: your outreach list |
A spreadsheet works fine. If you need hundreds of prompts across many engines, AI visibility tools automate the runs.
Step 4: turn results into work
- Not mentioned in category prompts: you lack third-party coverage. Get onto the cited pages; add news coverage with a press release and directory listings.
- Mentioned but described wrongly: fix the facts at the source. See when ChatGPT gets your brand wrong.
- Known with search, unknown without: normal for a new brand. The model's memory catches up as it is retrained on a web that mentions you.
- Competitor always first: read the cited pages and see why. Often one or two list articles drive the whole answer.
Questions people ask
Why does ChatGPT give me a different answer each time?
Answers are generated, not looked up, and search results vary. That is why you track the share of runs and prompts where you appear, rather than a single answer.
Does being logged in change results?
It can: memory, custom instructions and location influence answers. Test in temporary or logged-out sessions for a neutral baseline.