Where should your AI visibility prompts come from? Building a prompt set from real buyer questions
A visibility score is only as honest as its prompts. Take them from sales calls, tickets, Search Console and communities, weight them and refresh on a schedule.
The short answer
Build your AI visibility prompts from questions buyers have actually asked: sales call notes, support tickets, Search Console queries, site search and community threads. Weight them by buying stage and revenue, keep a fixed core for trend lines, and refresh the rest quarterly. Drop vanity prompts that name your brand or that no buyer would type.
Key takeaways
- A prompt set should be written from recorded buyer questions, not from the keywords a brand would like to rank for.
- Five sources cover most needs: sales calls, support tickets, Search Console queries, site search and community threads.
- Weight prompts by buying stage and the revenue behind them, so a high score reflects the questions that lead to deals.
- Keep a fixed core of prompts for trend lines and refresh the rest each quarter, recording every change.
- In Semrush's survey of US B2B professionals (March to April 2026), 92% of the 519 who use AI at work said AI had shaped their vendor shortlist.
In this article
Why the prompt set decides the score
An AI visibility score measures how often engines name you for a list of prompts, so the list decides what the score means. Choose prompts your brand already wins and the score flatters you. Choose prompts nobody asks and the score is precise about nothing. The only defensible list is one written from questions real buyers have asked.
Those questions now reach AI engines before they reach you. Semrush surveyed 622 US B2B professionals in March to April 2026 and reported on the 519 who use AI at work. Of those, 66% said they regularly use AI to research vendors and solutions.1 In the same group, 92% said AI had shaped their vendor shortlist.1 Semrush sells search and AI visibility tools.
Our view: the prompt set deserves more care than the dashboard. Our buyer's spec for an AI visibility score asks vendors to show their prompts; this post covers how to write your own, whether you buy a tool or not.
Five places real buyer questions live
Real buyer questions sit in five places most companies already own: sales calls, support tickets, Search Console, site search and community threads. Each gives a different slice of the buying journey, so draw from all five.
Sales calls and proposals
Sales notes and call recordings hold the shortlisting questions: "Which vendors handle multi-entity accounting?" or "What does implementation take for a firm our size?" Ask two or three sales people for the questions they hear in first calls, and read the questions sections of recent RFPs. These become your highest-weighted prompts.
Support tickets
Tickets hold the questions existing customers ask, which often match what prospects ask an engine before they buy: integrations, limits, pricing details and comparisons with what they used before. Group a quarter of tickets by theme and turn the top themes into prompts phrased as a prospect would ask them.
Search Console and site search
Search Console shows the queries people already use to find you. As Search Engine Journal reported, Google said its reports on impressions in AI Overviews and AI Mode had reached all websites worldwide as of 31 August 2026.2 Sites with few AI impressions may still not see one. Long, question-shaped queries are the best candidates, because they read like prompts. Your own site search adds questions people could not answer from your navigation. Our guide to Search Console, Bing and your own prompts explains what each of those reports counts, and why none of them replaces a prompt set.
Community threads
Reddit, industry forums, LinkedIn comments and Q&A sites show how buyers phrase questions to each other, including the doubts they would not raise with a salesperson. Collect phrasing, not just topics. "Is it worth it for a ten-person team?" is a prompt; "pricing" is a keyword.
| Source | Buying stage it covers | Typical weight |
|---|---|---|
| Sales calls and RFPs | Shortlist and comparison | High |
| Support tickets | Fit, limits and integrations | Medium |
| Search Console queries | Problem and category research | Medium |
| Site search | Gaps in your own content | Low |
| Community threads | Doubts and alternatives | Medium |
Illustrative. Our default weights; adjust them to where your deals actually come from.
How should you weight the prompts?
Weight prompts by buying stage and the revenue behind them, so the score moves when the questions that lead to deals move. A prompt such as "best ERP for a mid-sized manufacturer in Germany" deserves more weight than "what is an ERP", because only one of them produces a shortlist.
Shopping prompts lean the same way. In Profound's study of 200,000 commercial prompts during the 2025 holiday week (20 to 26 December 2025; geography not stated), 68.2% were product research and comparison.3 Profound sells AI visibility tracking. Our note on that study shows what those prompts looked like.
A simple scheme works. Tag each prompt with a stage (problem, category, shortlist, comparison) and a product line, then weight shortlist and comparison prompts at twice the others. Keep the weights in the same file as the prompts so anyone can check how the score was built.
From buyer questions to a prompt set
- Collect
Pull a quarter of questions from sales, support, Search Console, site search and communities.
- Rewrite
Phrase each as a buyer would type it to an engine, without your brand name.
- Tag and weight
Stage, product line and country; shortlist prompts weigh more.
- Split
A fixed core for trend lines, a rotating set for new questions.
- Refresh
Review the rotating set each quarter and log every change.
Which prompts are vanity prompts?
A vanity prompt is one your brand wins by construction. The commonest kind names you: "Is Northwind Digital good for analytics?" will mention Northwind Digital whatever the engine thinks. Others use your own product terms, which no buyer types, or a niche so narrow that you are the only vendor in it.
Keep a few branded prompts in a separate group, because they test accuracy rather than visibility. Our guide to fixing what ChatGPT gets wrong uses them that way. Never count them in the share-of-answer score.
Our view: if your first score is above what your sales team would believe, read the prompts before celebrating. A score built on vanity prompts tends to fall as soon as someone else writes the list.
How often should the prompt set change?
Change the rotating part quarterly and the core rarely, and record every change with its date. A trend line only means something if the prompts behind it stayed the same. When you must change a core prompt, run old and new side by side for one cycle so the break in the line is visible.
Refresh triggers are practical: a new product line, a new country, a competitor entering the category, or a question sales hears often that the set lacks. Our guide to picking a competitor set covers the matching rule for rivals, and our note on re-run frequency covers how often to run the set once it exists.
Our AI Visibility Audit uses 150 buyer prompts across six engines (ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI), running each prompt on three separate days. We use official APIs where offered and a licensed AI-answer data provider otherwise. We draft the prompts from a template for the client's industry and buying stages, the client approves or amends them at kick-off, and the full list is delivered with the results.
Sources
- Semrush, how AI shapes B2B buying: 622 US B2B professionals, Mar–Apr 2026; figures are of the 519 who use AI at work (Jul 2026)
- Search Engine Journal, Search Console AI reports rolled out worldwide (Aug 2026)
- Profound, holiday shopping prompts: 200,000 prompts, 20–26 Dec 2025; geography not stated (Jul 2026)
