Reading time: 10 minutes
Brand tracking prompts are fixed questions, repeated on the same AI engines, to count how many answers mention your brand. This prompt list gives 50 examples of our own, by question type and by industry, in blocks you can copy. It then shows how often to run them and how to log each answer.
In short
- A prompt list to track your brand covers four question types: category, problem, comparison and brand.
- The 50 prompts in this template are our own examples: 20 core prompts with fill-in fields in brackets, and 30 for six industries.
- The minimum protocol per brand is 50 prompts, 5 runs per engine and 4 engines, or about 1,000 answers a month.
- Each answer becomes a row in the spreadsheet, with engine, date, run and what the brand got: name, recommendation and citation.
What are brand tracking prompts, and how many do you need?
Brand tracking prompts are questions you write once and repeat the same way on ChatGPT, Gemini and other engines. So they measure how often your brand shows up, instead of relying on one isolated check. Per brand, the measurement guide recommends 50 prompts, 5 runs per engine and 4 engines a month.
If GEO is still new to your team, start with the guide to what GEO is. The prompts below are our own examples, written as a starting point. They don’t come from a survey of real searches or from market data. Therefore, test each one with your audience before you freeze the prompt list.

How do you build a prompt list to track your brand?
Build the prompt list around four question types, write each prompt the way a person would, and freeze it for a quarter. The four types come from the measurement guide. In this template, however, each one gets ready-made prompts.
Category shows whether your brand makes the list without being named in the question. Problem, in turn, shows whether it appears as a solution. Comparison sets your brand against competitors, while the brand type reveals what AI says about price and reputation.
Why vary the length and style of the prompts?
People don’t write to AI the way they type into Google. According to Google, the average AI Mode search in the US is triple the length of a traditional search query. Also, SparkToro asked volunteers to write one prompt for the same need: choosing headphones for a family member who travels.
The 142 prompts barely looked alike, with an average semantic similarity of 0.081. Still, Bose, Sony, Sennheiser and Apple appeared in 55% to 77% of the 994 answers. For SparkToro, the AI picked up the intent behind very different phrasings. The study does not name the volunteers’ country. It was done with Gumshoe.ai, an AI tracking startup. That startup employs the author’s research partner, and the author discloses the possible conflict of interest. SparkToro also sells software. The prompts were run on ChatGPT, Claude and Google’s AI results in late 2025, and volunteers were not required to be in the same country. So treat the result as a pattern to test in your own market, not a guarantee.
For that reason, write full sentences with context, not just keywords. We recommend two versions per intent, and each version counts as one prompt in the list:
SHORT
Best scheduling software for a medical practice.
WITH CONTEXT
I run a medical practice with 3 doctors in Austin
and want to reduce no-shows. Which scheduling
software do you recommend?
Which core brand tracking prompts should you use?
Use 20 core prompts, five per type. First, replace the fields in brackets with your brand’s details. Second, leave your brand out of the category and problem prompts, because they measure whether AI recalls it unprompted. The comparison and brand prompts, on the other hand, name it. Write each prompt in the language your customers use.
Category: does your brand make the list?
1. What is the best [CATEGORY] in [COUNTRY]?
2. Which [CATEGORY] do you recommend for
[AUDIENCE]?
3. What are the best [CATEGORY] companies
in [CITY]?
4. How do I choose a [CATEGORY]? List criteria
and options.
5. Which [CATEGORY] offer the best value
for money?
Problem: does your brand appear as a solution?
6. How do I solve [PROBLEM] in my company?
7. I have [PROBLEM]. What solution should
I hire?
8. What can I do to [GOAL] without spending
much?
9. What tool or service helps me [GOAL]?
10. I'm a [PROFILE] and I need to [GOAL].
Where do I start?
Comparison: how does your brand do against competitors?
11. [BRAND] or [COMPETITOR]: which should
I choose?
12. [BRAND] vs [COMPETITOR] for [AUDIENCE].
13. What is the difference between [BRAND] and
[COMPETITOR]?
14. What are the alternatives to [COMPETITOR]
in [COUNTRY]?
15. [CATEGORY]: [BRAND], [COMPETITOR 1]
or [COMPETITOR 2]?
Brand: what does AI say about your brand?
16. What is [BRAND] and what does it offer?
17. Is [BRAND] trustworthy? What do customers say?
18. How much does [BRAND] cost?
19. What are the most common complaints about
[BRAND]?
20. What are the alternatives to [BRAND]?
For prompts 18 and 19, the answer may show a wrong price or complaint. So log each incorrect claim, along with the source the AI cited.
Which prompts should you use in each industry?
Each industry gets 5 prompts that mix category, problem, comparison, price and local decision. So pick the block for your industry and add it to the 20 core prompts. The examples are ours and work as a starting point, so adjust the wording to your audience. For regulated industries, the guide to GEO by industry shows what changes.
Clinics and healthcare
21. Which [SPECIALTY] clinic is good in [CITY]?
22. How do I choose a [SPECIALTY] doctor who
takes [INSURANCE]?
23. How much does a [SPECIALTY] visit cost in
[CITY] without insurance?
24. [BRAND] clinic or [COMPETITOR] for
[PROCEDURE]?
25. What questions should I ask before booking
[PROCEDURE]?
B2B software
26. What is the best [FUNCTION] software for a
small company?
27. How do I automate [PROCESS] without hiring
more people?
28. [BRAND] or [COMPETITOR]: which integrates
better with [SYSTEM]?
29. How much does [FUNCTION] software cost?
30. What are the alternatives to [COMPETITOR]
with 24/7 customer support?
E-commerce
31. Where can I buy [PRODUCT] online with fast
shipping to [STATE]?
32. What is the best [PRODUCT] under $[AMOUNT]?
33. Is [BRAND] trustworthy for buying [PRODUCT]?
34. Which online [CATEGORY] store has the best
return policy?
35. Is [PRODUCT] worth it for [USE]?
Education
36. What is the best online [FIELD] program
for working adults?
37. Is [COURSE] worth it for a career change?
38. [BRAND] or [COMPETITOR] for studying [FIELD]?
39. How much does a [FIELD] certificate
program cost?
40. Which [FIELD] schools in [CITY] have good
reviews?
Real estate
41. What is the best real estate agent
in [CITY]?
42. How do I choose a neighborhood to buy
a home in [CITY]?
43. [BRAND] or [COMPETITOR] to sell my home?
44. How much does a property management
company charge to manage a rental?
45. Which home builders in [CITY] have good
reviews?
Professional services (legal and accounting)
46. Which CPA firm serves [SIZE] businesses
in [CITY]?
47. How do I choose a [AREA] attorney?
48. [BRAND] firm or [COMPETITOR] for [SERVICE]?
49. How much does a CPA charge to set up
an LLC?
50. What questions should I ask before hiring
a [AREA] attorney?
How do you get to 50 prompts for your brand?
Add the 20 core prompts, the 5 for your industry and 25 more derived ones. That gives you the 50 of the protocol. The 25 extras come from your audience’s real questions and from variations of the prompts you already have, because each variation tests a different angle.
First, pull real questions from Search Console, the sales team and support. To find long questions in Search Console, use the guide to regex in Search Console. The sub-questions that AI Mode asks behind a search also become prompts, with the guide to query fan-out.
To multiply the prompts you already have, change one element at a time and give each version a new code. For example, “best clinic in Austin” becomes “best clinic in East Austin”, and “for startups” becomes “for freelancers”. Use the list below as a script:
VARIATIONS OF ONE PROMPT
City: swap the city or the neighborhood.
Audience: swap who asks the question.
Budget: add a price range.
Phrasing: short version and version with context.
Competitor: one prompt per rival.
If your industry isn’t on the list, use the closest one and swap the vocabulary. Finally, take the prompt list to the sales team, who know the real questions customers ask.
How many times should you run the prompts to track your brand in AI?
Run each prompt 5 times per engine, on 4 engines, spread across the month. That is 50 prompts × 5 runs × 4 engines, or about 1,000 answers a month. At this volume, the margin of error is between ± 3 and ± 7 points.
Spread the runs out, because sources rotate. SISTRIX, which sells SEO software, tracked 82,619 prompts for 17 weeks. Google AI Mode replaced 56% of sources each week, and ChatGPT replaced 74%. The study covered six countries: Germany, the US, the UK, Italy, Spain and France.
Also, the margin of error depends on the total number of answers. These numbers come from the measurement guide, at 95% confidence:
| Collection | Answers per month | Margin of error |
|---|---|---|
| 10 prompts, 3 runs, 1 engine | 30 | ± 18 to ± 31 points |
| 50 prompts, 5 runs, 1 engine | 250 | ± 6 to ± 14 points |
| 50 prompts, 5 runs, 4 engines | 1,000 | ± 3 to ± 7 points |
A thousand answers by hand is a lot of work. So start with one engine, which yields 250 answers, and add the others later. With 1,000 answers, treat only a change above 4 to 10 points as real, confirmed the following month. Once the volume passes a few hundred answers a month, move the collection to a tool.
Next, on every run, fix everything that changes the answer, except time:
COLLECTION RULES
1. New chat on every run.
2. Temporary chat on ChatGPT and a private
window on the other engines.
3. Same country and language on every run,
such as the US and English.
4. Same prompt text, no edits.
5. Save the full answer and the date.
Finally, the measurement guide details each rule in its section on how to control collection.
How do you log the answers to the prompts?
First, log one row per run in a spreadsheet, with the full answer pasted in. Each prompt gets a fixed code, such as P07, that doesn’t change between months. That way, the spreadsheet compares the same prompt over time. So brand tracking prompts only pay off if every answer is saved.
| Field | What to log | Example |
|---|---|---|
| Code and type | Fixed ID, type and industry | P07, comparison, clinics |
| Date and engine | Collection day and engine | Nov 12, Gemini |
| Brand | Named and recommended | Yes, no |
| Site and competitors | Cited domain and rivals | brand.com, A and B |
| Sentiment | Positive, neutral or negative | Neutral |
Copy the list below into the first row of the spreadsheet, one column per item:
prompt_id
type
industry
date
engine
run
brand_named
brand_recommended
site_cited
competitors_named
domains_cited
sentiment
full_answer
Count it as a recommendation when the answer points to the brand as an option, in a list or with a verb like “consider”. A passing mention, however, counts only for presence. Once the spreadsheet is full, the rates come from simple math, such as answers with the brand ÷ total × 100. The guide to AI share of voice shows the math with competitors.
What does each type of prompt show about the brand?
Compare the rates by question type, not just the average of the prompt list. Each type shows one part of your visibility, so reading them separately helps you pick the next action. The readings below are hypotheses to investigate, not a diagnosis.
- High brand, low category: AI knows the name, but may not link it to the category when nobody names it.
- High category, low comparison: the brand makes the list, but may lose to competitors when the question is direct.
- Low problem: the brand may not appear as a solution, since AI may not link it to the customer’s pain.
- Brand with errors: AI repeats a wrong price or fact. The fix can start at a source outside your site that the AI used.
In addition, each type gathers fewer answers than the whole list. So the margin of error per type is larger than for the whole list, and only a big change, confirmed the following month, deserves action.
When should you change the prompt list?
Change the prompt list every quarter, not every month. Because AI already varies on its own, changing the prompts adds a second source of variation to the dashboard. So, when you review, keep most of the prompts, note the date of the change and compare only the ones that stayed. This is our own recommendation, in line with the measurement guide.
New competitors, new cities and the questions the sales team started to hear all go into the review. Prompts for products the brand no longer sells leave the list. Each new prompt gets a new code, and the old ones keep their history. That way, your brand tracking prompts follow the business without losing the comparison with earlier months.
Frequently asked questions
Yes, as a draft. Then review it and add sentences with context, because at SparkToro, 142 prompts written by volunteers for the same need had an average similarity of 0.081. Next, freeze the prompt list and log the answers in the same spreadsheet. That way, the prompt list no longer depends on who wrote it.
No. In the category and problem prompts, the brand stays out, because they measure whether AI recalls it unprompted. Only the comparison and brand prompts name it. So the prompt list mixes both types, and each one answers a different question about the brand.
You can for the 20 core prompts, by swapping the fields in brackets. Each client, however, needs the 5 prompts for its industry, its cities, its competitors and the questions of its own audience. Without that, the prompt list measures the industry, not the brand. Finally, freeze one prompt list per client.



