AI visibility
How do you build an AI visibility prompt list?
Updated October 2026 · Manoj Surya
Quick answer
Build the list from questions buyers ask before they know your name: category questions, problem questions, 'alternative to' questions for your main rivals, and head-to-head comparisons. Start with 10 to 20 prompts, word them as full questions, and do not change them between runs. MarketHQ tracks 10, 60 or 120 questions on Lite, Startup and Growth.
Key points
- Leave your own brand name out of the prompts; people who already know you are not the buyers you need to win.
- Mix four types: category, problem, alternative-to and comparison questions.
- Keep the wording identical every week, or the trend measures your edits instead of the engines.
- MarketHQ seeds a starting list from your category, competitors and keywords when you sign up, and you can edit it in settings.
- MarketHQ asks every tracked question in six AI engines each week on paid plans.
Founder of MarketHQ. Building market intelligence for companies that sell to developers.
Every AI visibility number depends on the questions behind it. A list of vanity prompts gives you a flattering score that predicts nothing, while a list of real buyer questions shows you where deals are being decided. This guide covers where to find those questions, how to word them, and how many you need.
What makes a good AI visibility prompt?
A good prompt is a full question a real buyer would type before they have heard of you, worded the way they would word it.
Search keywords are too short for this job. Nobody types 'invoicing software' into ChatGPT. They type 'what is the best invoicing software for a five-person agency that bills in two currencies'. The longer question pulls a different answer, and that answer is the one your buyer reads.
Leave your brand name out. 'Is Acme good for invoicing' will almost always mention Acme, so it inflates your score and tells you nothing. The prompts that matter are the ones where the assistant has to decide who to name.
Which types of prompts should the list include?
Cover four intents in roughly equal parts: category, problem, alternative-to and comparison.
Each type catches a different stage of buying. Category questions show whether you are on the default shortlist. Problem questions show whether assistants connect you to the job you do. Alternative questions show whether you are offered to people leaving a competitor, and comparison questions show how you are described once you are already in the frame.
| Type | Example prompt | What it tells you |
|---|---|---|
| Category | What are the best invoicing tools for small agencies? | Whether you make the default shortlist |
| Problem | How do I stop chasing late client payments? | Whether assistants link you to the job |
| Alternative to | What is a cheaper alternative to [main competitor]? | Whether you are offered to switchers |
| Comparison | [Competitor A] vs [Competitor B] for freelancers | Whether you get added to a head-to-head |
Where do you find the questions buyers actually ask?
Start with your own sales and support conversations, then add search data and community threads.
The best prompts are usually written down somewhere in your company already. They just are not filed as prompts yet.
- Sales calls and demo notes: the question a prospect asked in the first five minutes, and the competitors they named.
- Support tickets and onboarding emails: 'how do I do X' questions that a non-customer would also ask.
- Google Search Console: filter your queries to ones starting with how, what, which or best. Many are already phrased as questions.
- Community threads: Reddit, Hacker News and industry forums where people ask for recommendations in your category.
- Lost-deal notes: 'we went with X because of Y' becomes 'what is the best tool for Y'.
How many prompts do you need to track?
Start with 10 to 20, and add more only when you are acting on what the first set tells you.
Ten well-chosen questions across six engines gives you 60 answers a week. That is enough to see where you are missing and who is named instead. A list of 200 questions nobody reads is worth less than 20 you act on.
Grow the list when you add a product line, enter a new market, or see a competitor turn up in answers you did not expect. MarketHQ tracks 10 questions on Lite ($29 a month), 60 on Startup ($99) and 120 on Growth ($199), so the plan limit also pushes you to pick the questions that matter.
How should you word and maintain the list over time?
Freeze the wording, retire questions instead of editing them, and review the list once a quarter.
If you rewrite a question halfway through a quarter, the trend line stops measuring the engines and starts measuring your edit. When a question needs to change, retire the old one and add the new one as a separate row, so each keeps a clean history.
MarketHQ follows the same rule: the exact text you see in settings is the text each engine receives, with no silent rewording between weeks. At sign-up it seeds a starting list from your category, competitors and keywords, mixing discovery, how-to, comparison and alternative questions. Treat that as a first draft and replace anything that does not sound like your buyer.
- Write each prompt as a full sentence with the context a buyer would add, such as team size, industry or budget.
- Tag each prompt with its type, so you can spot patterns like losing every comparison question.
- Review the list quarterly. Weekly edits make the trend unreadable.
What does a first prompt list look like?
Illustrative example: a payroll startup turns demo notes and Search Console queries into 12 questions.
Illustrative example: a payroll startup for restaurants pulls the opening question from ten recent demo notes and filters Search Console for question-shaped queries. It ends up with three category questions ('best payroll software for restaurants'), three problem questions ('how do I pay tipped employees correctly'), three alternative questions naming its two largest competitors, and three comparisons. In week one, assistants name it in some category answers and in none of the problem answers. That tells the team its help content is not being connected to the product, so the first fix is a clear guide on paying tipped staff that names the product in the steps.
What can a prompt list not tell you?
It shows how engines answer your chosen questions, not every question buyers ask or what each user sees.
Your list is a sample. A buyer can phrase the question differently, ask a follow-up, or use an assistant with memory and web search switched on. Treat the share you measure as a trend to act on, and check it against what your sales team hears on calls.
FAQ
- How many prompts should I track for AI visibility?
- Start with MarketHQ Lite: 10 buyer questions asked in six AI engines every week for $29 a month, which is 60 answers a week. Move to 60 questions on Startup or 120 on Growth when you add product lines or markets.
- Should my brand name be in the prompts?
- No. A prompt that names you will usually mention you, which inflates the score. Track questions a buyer asks before they know your name, plus alternative-to questions for your competitors.
- Can I change prompts after I start tracking?
- Yes. Retire the old prompt and add a new one instead of editing it in place. Editing breaks the history, so you can no longer tell whether the engines changed or your wording did.
- Is there a way to see which prompts mention my brand?
- MarketHQ shows, for each tracked question and each engine, whether the answer named you, who it named instead and which links it cited, on paid plans. For a quick one-off reading, the free LLM mention checker asks ChatGPT and DeepSeek two fixed prompts.
Related
- See which prompts mention your brand — reading the per-question results
- Track brand mentions in AI search — the weekly tracking routine
- Gap analysis on AI answers — what to do with the questions you lose
- AI visibility use case — how MarketHQ runs your list every week