Your AI visibility numbers are only as good as the prompts behind them. Track ten prompts that all contain your product’s niche and you will look great. Track ten generic prompts like "best analytics tool" and you will look invisible next to companies with ten years of press. Neither tells you much about your buyers.
This guide shows how to choose a prompt set that reflects how buyers actually ask, how many prompts you need, why you should keep them stable, and how to tag them. It ends with a complete example set for a fictional product.
What makes a good tracking prompt
A prompt is a question you send to an AI assistant, in the same words, every time you check. A good one has four properties:
- A buyer would type it. It sounds like a person with a problem, not like a keyword. "What’s the best product analytics tool for a seed-stage B2B startup?" beats "product analytics software".
- It doesn’t name you. Prompts that contain your brand will almost always mention you and inflate your numbers. Track those separately if at all.
- Winning it would matter. If you appeared in this answer, could it lead to a signup? A definition question like "what is product analytics?" rarely produces a buyer.
- It stays the same. You will run it for months. Write it once, carefully.
The five prompt types
Cover several types so one kind of question doesn’t dominate your numbers.
| Type | Shape | Example | Why track it |
|---|---|---|---|
| Category discovery | "best X for Y" | best product analytics tool for a seed-stage startup | This is where shortlists are built. The most valuable type for most founders. |
| Comparison | "X vs Y" | Mixpanel vs PostHog for a small B2B SaaS | Buyers close to a decision. Shows whether you get pulled into the comparison. |
| Alternatives | "X alternatives" | Amplitude alternatives for a small team | People unhappy with an incumbent. Often the easiest place for a small product to appear. |
| Problem-led | "how do I…" | how do I see which features new users actually use? | Buyers who don’t know the category name yet. Tests whether you are linked to the problem. |
| Use-case | "tool for [situation]" | analytics for a two-person SaaS with no data engineer | Matches your positioning. Small products can win these against big ones. |
You can also add one or two brand prompts ("what is Acme Analytics?", "is Acme Analytics good for B2B SaaS?") to check how assistants describe you. Tag them separately and leave them out of headline numbers.
Where to find real buyer prompts
- Ask customers. On your signup form or in onboarding calls, ask how they found you and, if it was an AI assistant, what they asked. Their words are the best prompts you will get.
- Sales and support notes. The questions prospects ask before buying ("does it work without a data engineer?") translate directly into problem-led prompts.
- Search terms. Your Search Console queries and keyword research show demand. Rewrite them as full questions. "crm small agency" becomes "what CRM should a five-person agency use?"
- Communities. Threads where people ask for recommendations in your category show the phrasing and constraints buyers mention.
- Competitor pages. The alternatives and comparison pages your competitors publish tell you which matchups they think matter.
How many prompts to track
| Stage | Prompt count | Why |
|---|---|---|
| First manual check | 10 to 20 | Enough to see patterns. Small enough to run by hand across a few assistants. |
| Ongoing tracking | 20 to 50 | Covers each prompt type with several prompts, so one answer changing doesn’t swing everything. |
| Several products or segments | 20 to 50 per segment | Keep segments separate with tags so their numbers don’t blur together. |
More prompts is not automatically better. Each prompt runs on each engine selected for it, so 50 prompts across four engines is 200 checks per run. Start with a set you will actually read, then add prompts where you see gaps.
A rough split for a 25-prompt set: 8 category discovery, 5 comparison, 4 alternatives, 4 problem-led and 4 use-case. Adjust toward the types your buyers use most.
Keep prompts stable
AI answers are sensitive to wording. "Best analytics tool for startups" and "best analytics tools for a startup" can produce different names. If you tweak prompts every week, your trend line mixes different questions and means nothing.
- Write each prompt carefully once, then freeze it.
- If a prompt is wrong, retire it and add a new one. Treat the new one as a new baseline.
- Keep a short log of what changed and when.
- Add new prompts freely. Adding doesn’t break the history of existing ones, though it does change any number that averages across all prompts.
Tag your prompts
Tags let you slice results without keeping separate lists. Useful tags:
- Type: discovery, comparison, alternatives, problem, use-case, brand.
- Segment: the audience in the prompt, such as seed-stage, agency or ecommerce.
- Competitor: for comparison and alternatives prompts, the competitor named.
- Priority: a "core" tag for the prompts you would report on if you could only pick ten.
With tags you can answer questions like "how do we do on alternatives prompts?" or "are we visible to agencies at all?" without building a new prompt set.
An example prompt set
Acme Analytics is a fictional product analytics tool for small B2B SaaS teams. Its main competitors are Mixpanel, PostHog and Amplitude. Here is a 20-prompt starting set.
| # | Prompt | Tags |
|---|---|---|
| 1 | best product analytics tool for a seed-stage B2B startup | discovery, seed-stage, core |
| 2 | what analytics tool should a small SaaS team use to track activation? | discovery, core |
| 3 | best product analytics tools for startups on a small budget | discovery, seed-stage |
| 4 | which product analytics tools are easiest to set up? | discovery |
| 5 | best analytics tool for a product-led SaaS | discovery, core |
| 6 | top product analytics software for B2B companies | discovery |
| 7 | Mixpanel vs PostHog for a small B2B SaaS | comparison, mixpanel, posthog, core |
| 8 | PostHog vs Amplitude for an early-stage startup | comparison, posthog, amplitude |
| 9 | Mixpanel vs Amplitude: which is better for a small team? | comparison, mixpanel, amplitude |
| 10 | Mixpanel alternatives for a small team | alternatives, mixpanel, core |
| 11 | Amplitude alternatives that are simpler to use | alternatives, amplitude |
| 12 | PostHog alternatives without self-hosting | alternatives, posthog |
| 13 | how do I see which features new users actually use? | problem, core |
| 14 | how can I measure activation in my SaaS without a data engineer? | problem |
| 15 | how do I find where users drop off during onboarding? | problem |
| 16 | analytics for a two-person SaaS with no data team | use-case, seed-stage, core |
| 17 | tool to track feature adoption in a B2B app | use-case |
| 18 | product analytics for a SaaS that sells to enterprises | use-case |
| 19 | what is Acme Analytics? | brand |
| 20 | is Acme Analytics a good choice for B2B SaaS? | brand |
Notice what this set does: it never mentions Acme outside the two brand prompts, it covers all five types, it names each competitor in at least one comparison and one alternatives prompt, and seven prompts are tagged "core" for a quick headline view.
Common mistakes
- Only tracking head terms. "Best CRM" is dominated by the biggest brands. Add the specific situations you actually serve.
- Writing prompts after looking at results. Choosing prompts you already win makes the numbers meaningless. Pick the set first.
- Too many near-duplicates. Five phrasings of the same question count one intent five times. Keep one and vary the type instead.
- Changing wording silently. Every edit is a new baseline. Log it.
- Forgetting the problem-led prompts. Buyers who don’t know your category’s name are often the ones a small product can win.
Putting the set to work
Run the set by hand with the method in how to track brand mentions in ChatGPT, or try the free AI visibility checker for a structured first pass. To run it on a schedule across ChatGPT, Claude, Perplexity and Google AI Overviews, prompt tracking in AEOmeter lets you paste or import a list (CSV or plain text, with a preview), tag prompts, pause them and see each prompt’s history. When you add a brand, AEOmeter also suggests starter prompts you can edit before saving.
FAQ
Should prompts be questions or keywords?
Questions, written the way a person types into a chat box. Assistants handle full sentences, and the constraints in a question ("for a two-person team") change which brands get named.
Should I include my brand name in prompts?
Only in a separate, tagged set of brand prompts used to check how you are described. Buyer prompts should not contain your name.
How often should I change my prompt set?
Rarely. Add prompts when you spot new buyer questions or competitors, and retire prompts that no longer fit. Avoid editing the wording of prompts you are trending.
Should I use the same prompts for every AI assistant?
Yes. Using identical prompts lets you compare engines. Just compare each engine’s trend with itself, since they behave differently.
What about follow-up questions?
Real buyers often ask follow-ups in the same chat. Single prompts are easier to track consistently, so start there and read a few multi-turn conversations by hand when you want more depth.
