How to Track Your Brand in ChatGPT

TL;DR: Pick 20 questions your buyers ask, run each one several times, and record whether you were named, where, how you were described and which sources were cited. Then put the results in a spreadsheet and read the patterns.

Pick 20 prompts

Start from things you can check. Search Console queries you already get impressions for are real demand in your buyers' words, and the head terms for your category from the search results are a good second source. If you'd rather not write the set yourself, the prompt set generator builds one for you.

Replace [category] with how your buyers describe what you sell, [your brand] with your name and [rival] with your closest competitor. Keep the wording your buyers use, not your internal positioning.

Category and discovery (5)

1. What is the best [category] in 2026?
2. What are the top [category] options for a small team?
3. I need [the outcome your product delivers]. What should I use?
4. What is the best free [category]?
5. Which [category] is best for [your primary customer type]?

Comparison and alternatives (4)

6. [your brand] vs [rival], which is better?
7. What are the best alternatives to [rival]?
8. Is [your brand] any good?
9. What do people say about [your brand]?

Pricing (2)

10. How much does [category] cost?
11. What is the cheapest [category] that is actually good?

Problem-aware (5)

12. Why is [the problem your product solves] happening?
13. How do I [the job your buyer is trying to do]?
14. What is the easiest way to [the outcome]?
15. I tried [the common DIY approach] and it did not work. What now?
16. Is [category] worth paying for, or can I do it myself?

Objection and risk (4)

17. What are the problems with [category]?
18. Is [category] a scam?
19. What should I watch out for when buying [category]?
20. [category] vs doing it manually, which is better?

Most people skip the problem-aware block. In Peak Answer's own measurement, the comparison and pricing style questions produced citation sets made up almost entirely of competitors and comparison sites, which are pages you can't get a link from. The problem-aware questions produce a quite different source list.

Questions 6 to 9 name your brand, so keep them as their own block and don't count them toward your presence rate.

Run each prompt more than once

Ask ChatGPT the same question 5 times and you will often get different lists.

Run every prompt at least 3 to 5 times and record the average. A single run is a small sample size, so you can't tell a real improvement from a random change in the answer.

Use a logged-out session where you can. There's no memory or custom instructions, so results are comparable over time and between brands. It's what a new visitor sees.

Also note which model answered each run. When OpenAI changes the underlying model, scores can shift for reasons that have nothing to do with you.

What to record

Set up a spreadsheet with one row per prompt, per engine, per run. The columns:

• Date. Answers drift, so every result needs a date.
• Prompt number. 1 to 20.
• Block. Category, comparison, pricing, problem or objection.
• Engine. ChatGPT, Claude, Perplexity or Google.
• Named. Yes or no. This gives you your presence rate.
• Position. First, middle, last or n/a. Third beats seventh.
• Others named. Every other brand in the answer, in order. This is your real competitive set.
• Cited. Yes or no. If the answer showed sources, the engine searched. This is the most useful column.
• Link. Whether anyone got a clickable link.
• Sources. The domains cited.
• Sentiment. Positive, neutral, negative or absent. Being named badly is worse than being absent.
• Notes. Anything that surprised you.

Sort by Cited and you can see how often the engine searches. Pivot on Sources and you have a list of the pages that decide your answers.

How to read it

Start with three counts.

• Presence rate. How many answers named you. Under 20% is normal for a brand that has never worked on this. Over 60% means your problem is position, not presence.
• Retrieval rate. How many answers showed citations. Mostly cited (over 60%) means the engines are searching, so content and placement work can move results in weeks. Mostly uncited (under 30%) means they are answering from memory, which is slower and comes from reviews, directories and a consistent description over months.
• Source tally. Count every domain in the Sources column. Usually 3 to 5 domains account for most of the citations. Open the top 5 and search for your brand name.

Then match what you see to the table.

Other engines

Run the same 20 prompts on each engine your buyers use. They disagree, and the disagreement is useful.

• ChatGPT. Varies the most between asks, and searches more readily on commercial and current questions than on definitional ones.
• Perplexity. Searches on nearly everything and cites heavily, so it's the best engine for collecting a source list.
• Claude. Searches less often than Perplexity and more often than people assume. Check for citations instead of going by tone.
• Google AI Overviews. The most sensitive to location and personalization, so use an incognito window.
• Gemini. Closest to Google, so your traditional search position matters more on this one.

When to stop doing it by hand

Do the manual version once, properly. Reading an answer where 3 competitors are named and you aren't is more motivating than any dashboard. But 5 runs across 20 prompts every week is 100 queries a week, and nobody keeps that up.

Once you've run it twice and want to know what changed, switch to an AEO tracker. It asks the same questions on a schedule across the engines and records the sources for you. If you only want a quick first look before building the spreadsheet, the free audit does that in about a minute.

Related study: Who Does AI Cite as Questions Get More Specific?