Home / Resources / How to build an AI prompt panel
GEO · MeasurementHow to build an AI prompt panel to test your brand’s visibility
You cannot improve what you do not measure, and AI answers change from run to run. A prompt panel gives you a repeatable baseline. Here is how to build and score one.
Key takeaway: A prompt panel is a fixed list of real buyer questions you run in each AI assistant on a schedule. Score whether you are mentioned, whether you are recommended and whether the description is accurate. Because answers vary run to run, judge trends over repeated runs rather than any single answer.
1. Choose 20 to 40 questions
Write the questions your real buyers ask, in their words. Mix four types:
- Category: “best [service] for [type of customer]”.
- Comparison: “[you] vs [competitor]” and “alternatives to [competitor]”.
- Problem: “how do I fix [problem you solve]”.
- Brand: “what is [your company]” and “is [your company] any good”.
Include a share of unbranded questions. If you only test questions that contain your name, you will overestimate your visibility.
2. Run them consistently
- Use the same questions each time, in a fresh session without personalisation where possible.
- Test ChatGPT, Gemini, Perplexity and Google, because they behave differently. Add Claude if your buyers use it.
- Note the date, the tool and whether web search was on.
- Run important questions more than once, since answers vary.
3. Score each answer
| Measure | What to record |
|---|---|
| Mentioned | Is your brand named at all? |
| Recommended | Is it suggested as an option, and in what position? |
| Accurate | Are the facts about you right: services, location, pricing, claims? |
| Cited | Does the answer link to your site or a page about you? |
| Competitors | Who is named instead, and why might they be? |
4. Read the results honestly
- Look at unbranded questions first. They show whether buyers who do not know you find you.
- Treat a single appearance or absence as noise. Look at the rate across many runs.
- Fix inaccurate answers at the source: your own pages, profiles and the third-party pages the assistant cites.
5. Repeat and compare
Re-run the panel monthly, log changes you made and compare against the baseline. Expect technical fixes to show up faster than earned reviews and mentions. See how to track AI search visibility and the AI visibility checklist for what to fix.
To skip the setup, the free AI Rank Tracker and AI Visibility Test Panel build a starter panel from your brand and category.
Frequently asked questions
What is a prompt panel?
A fixed list of real buyer questions that you run in AI assistants on a schedule to measure how often and how accurately your brand appears.
How many questions do I need?
Twenty to forty is enough for a useful baseline. More questions give a steadier picture but take longer to run and score.
Why do answers change between runs?
AI assistants generate answers probabilistically and may use different sources, so single answers are noisy. Judge trends across repeated runs.
Is there a free tool for this?
Reckona AI offers a free AI Rank Tracker and AI Visibility Test Panel that build a starter panel for your brand and category, which you then run yourself.
Want this done for your brand?
Start with a free audit: a written scope and a baseline of how AI assistants describe your brand today.
Get my free audit →