Checking whether ChatGPT, Gemini and Perplexity mention your brand takes about an hour a month and a spreadsheet. You write a fixed set of buyer prompts, run each one in every assistant, and record what comes back. This is the hands-on companion to Measuring AI visibility without fooling yourself, which explains why the method works.
Can I just ask each assistant about my brand once?
You can, but one answer tells you very little. Answers vary by model, date, location, phrasing and personalisation, and there is random variation even when you repeat the same prompt. A single run is an anecdote. A fixed prompt set, run the same way each month, gives you a trend you can defend.
What do the three assistants actually show?
Each one can show where an answer came from, which matters because the sources are half of what you are checking.
- ChatGPT search gives answers with links to relevant web sources. It decides on its own whether to search, and you can choose it manually with the web search icon.
- Perplexity says it searches the internet in real time and that each answer includes numbered citations linking to the original sources.
- Gemini sometimes shows sources and related content. Its own help page notes that not all responses include links or sources.
Features change, so confirm what you see on screen at the time of each run.
How do I check my brand step by step?
- Build 20 to 40 buyer prompts. Write them the way a customer would, not the way a marketer would. Group them by intent: category ("best [category] in [location]"), comparison, problem, brand and buying. Aim for four to eight per group.
- Freeze the list. Save the exact wording. Changing prompts between months ruins the comparison.
- Run each prompt logged out or in a private window. This reduces the effect of your own history and saved memory. Turn web search on where there is a switch, so you can see cited sources. Note your location, because it can shape local answers.
- Run it in all three assistants. If you can, repeat each prompt three times on different days. Fewer runs and you are measuring one afternoon.
- Record the result in the template below. Copy the answer text for any claim made about you.
- Mark mentioned and recommended separately. Mentioned means your name appears. Recommended means the assistant points the reader towards you, for example as its top pick or in a shortlist it endorses. Mentioned without recommended is a positioning problem.
- Check the claims. Is the description of your services, location and pricing correct? Wrong claims are an entity problem and often point to an outdated third-party page.
- Note competitors and cited sources. The sites cited in answers you lose are where to look for gaps.
- Repeat monthly, and after a launch or rebrand. Weekly runs mostly measure noise.
What should my tracking template look like?
A plain spreadsheet is enough. Paste this into a sheet or document and add a row per run.
| Date | Assistant | Prompt | Mentioned | Recommended | Position | Competitors named | Source cited | Notes |
|---|---|---|---|---|---|---|---|---|
| 2026-12-01 | Perplexity | Best [category] in [location] | Yes | No | 3 of 5 | [Competitor A], [Competitor B] | [Directory URL] | Description of services was out of date |
| 2026-12-01 | ChatGPT | Best [category] in [location] | No | No | n/a | [Competitor A] | [Review site URL] | Not named at all |
Use Yes or No for the two main columns so you can count them. Position is the place in any list, or n/a if you are not in one. Keep the full wording of any inaccurate claim in Notes.
Which prompts should I start with?
These ten patterns cover most buying journeys. Replace the [brackets] with your own category, location and competitors.
- Best [category] in [location]?
- Who are the top [category] providers for [type of customer]?
- [Your brand] vs [competitor]: which is better for [use case]?
- What are the alternatives to [competitor]?
- Who should I hire for [job] in [location]?
- How do I fix [problem your service solves]?
- What is [your brand] known for?
- Is [your brand] any good? What do customers say?
- What should I look for when choosing a [category] supplier?
- Recommend a [category] under [budget] for [situation].
Prompts 1, 2 and 5 test discovery. Prompts 3 and 4 test competitor displacement. Prompts 7 and 8 test whether the assistant understands your brand at all.
How do I read the results honestly?
Look at rates, not single instances. "We were mentioned in 11 of 30 prompts, up from 8" is a finding. "ChatGPT named us once" is not. Keep the three assistants separate, because they behave differently and cite different sources.
Be upfront about the limits. Your private-window run is not what every customer sees, and results shift with location and personalisation. Treat the output as a sample, not a ranking. Our glossary defines terms such as entity, citation and answer engine if any of this is new.
Is there a faster way to get a first read?
Yes. The free AI Visibility Checker runs a four-prompt snapshot, which is useful for a quick look before you commit to the full 20 to 40 prompt set. It is a starting point, not a replacement for repeated monthly runs.
In short
- Build a fixed set of 20 to 40 buyer prompts grouped by intent, and run them logged out with web search on in ChatGPT, Gemini and Perplexity.
- Record mentioned and recommended separately, plus competitors, cited sources and whether the claims about you are accurate.
- Repeat monthly, read trends rather than single answers, and accept that results vary by model, date, location and personalisation.