Why the Prompt Set Is the Measurement
An AI visibility score is only as good as the questions behind it. Track ten prompts that all contain your brand name and you will score well and learn nothing. Track prompts your buyers never ask and the trend line will not connect to revenue. This template gives you a structure for choosing prompts, a starter allocation of 25, and a sheet layout for recording results. It is written so a team can fill it in during one working session.
Step 1: Use Five Prompt Types
Every prompt in the library should belong to one of five types. Each type answers a different question about your visibility.
| Type | What it tests | Pattern | Brand name in prompt |
|---|---|---|---|
| Category | Whether you are named when someone asks for options | "What are the best [category] tools for [audience]?" | No |
| Problem | Whether you are suggested as a solution before the buyer knows the category | "How do I [job to be done] without [pain]?" | No |
| Comparison | How you are positioned against a named competitor | "[Brand] vs [Competitor] for [use case]" | Yes |
| Brand | Whether the facts about you are right | "What does [Brand] do and how much does it cost?" | Yes |
| Purchase | Whether you survive the final shortlist | "Which [category] should a [size] [industry] company buy?" | No |
The three unbranded types matter most. They show whether an assistant brings you up without being prompted, which is the situation a new buyer is in.
Step 2: Allocate 25 Prompts
Twenty-five prompts is enough to cover one brand in one market with a stable trend. Use this split as a starting point and adjust it to where your pipeline comes from.
| Type | Prompts | Share | How to choose them |
|---|---|---|---|
| Category | 8 | 32% | Your main category phrased four ways, plus four audience or industry variants |
| Problem | 6 | 24% | The six problems that appear most often in sales call notes |
| Comparison | 5 | 20% | One for each of your five most common competitors in deals |
| Brand | 3 | 12% | What you do, what you cost, and who you are for |
| Purchase | 3 | 12% | One each for your three best-fit buyer profiles |
| Total | 25 | 100% |
Step 3: Write Prompts the Way Buyers Talk
- Take wording from real sources. Sales call transcripts, support tickets, site search logs, and the questions people ask in community forums. Do not write prompts from your own positioning document.
- Keep them long. People write full sentences to an assistant, with context about their company and constraints. A prompt of 15 to 25 words is normal.
- Include the constraint. Budget, team size, region, integration, or compliance need. Constraints change which brands an assistant names.
- One question per prompt. A prompt that asks two things gives an answer you cannot score cleanly.
- Leave your brand out of unbranded prompts. That includes product names and taglines.
Step 4: Fill In the Library
Copy this layout into a sheet. One row per prompt.
| ID | Type | Prompt | Buyer profile | Funnel stage | Source of wording | Competitors expected | Owner |
|---|---|---|---|---|---|---|---|
| C-01 | Category | [Write the prompt] | [Who asks this] | Consideration | [Call notes, forum, search log] | [Names] | [Name] |
| P-01 | Problem | [Write the prompt] | [Who asks this] | Awareness | [Source] | [Names] | [Name] |
| V-01 | Comparison | [Write the prompt] | [Who asks this] | Decision | [Source] | [Names] | [Name] |
| B-01 | Brand | [Write the prompt] | [Who asks this] | Decision | [Source] | [Names] | [Name] |
| X-01 | Purchase | [Write the prompt] | [Who asks this] | Decision | [Source] | [Names] | [Name] |
Step 5: Record Results the Same Way Every Time
For each prompt, on each assistant, on each run, record these fields. Run every prompt more than once, because answers vary between runs.
| Field | What to record |
|---|---|
| Date, assistant, model, plan | Answers differ by model and by free or paid plan, so note all four |
| Mentioned | Yes or no |
| Position | Where in the answer your brand first appears |
| Competitors named | Every other brand in the answer |
| Accuracy | Any wrong fact about price, features, or company |
| Sources cited | The URLs the assistant linked |
| Sentiment | Positive, neutral, or negative framing |
Step 6: Review the Library Every Quarter
- Retire prompts that never change. A branded prompt you win every run for three months is not telling you anything.
- Add prompts for new launches and new competitors.
- Keep at least two thirds of the set stable. If you replace everything, you lose the trend.
- Re-check wording against fresh call notes. The way buyers describe a problem drifts.
Common Mistakes
- Too many branded prompts. They inflate the score and hide the gap in unprompted recommendations.
- Keyword-style prompts. "best crm software" is a search query. People do not talk to an assistant that way.
- One run per prompt. A single answer is a sample of one.
- Changing the set every month. Trends need the same questions asked repeatedly.
How Presenc AI Helps
Presenc AI runs your prompt library across the major AI assistants on a schedule and records mentions, position, competitors, and cited sources for each run. The Starter plan tracks 25 prompts, which is the allocation this template is built around.