/
Guides
/
3 min read
Prompt research for AEO: how to build a tracking set that reflects real buyers
Keyword lists make poor prompt sets. Here’s the method we use to build 150–300 prompts that match how buyers actually ask AI engines for help.
Sofia Marchetti
Head of content
Most teams start AEO by pasting their keyword list into a tracking tool. The result looks thorough and measures the wrong thing, because buyers don’t type keywords into ChatGPT; they describe their situation and ask for a recommendation. A good prompt set reads more like a transcript of your sales calls. Here is the method we use with customers to build one in about a week.
Collect questions in your buyers’ words
Start where buyers use their own language: sales call notes, support tickets, community threads and the free-text field on your demo form. Pull 300–500 raw questions and resist the urge to clean them up. The odd phrasing is the point.
Group by intent, not topic
Sort the questions into four groups: category, problem, comparison and brand. “How do I stop spreadsheets breaking our month-end close” is a problem prompt, even though it never names a product category. Problem prompts are where buyers who don’t know your category yet will first meet you.
Rewrite each one as a full prompt
Add the context a real buyer would include: company size, industry, budget and current tools. Answers change with that context, so your tracking should include it too.
Weak: “CRM for startups”
Better: “What CRM should a 15-person B2B startup use if we already run HubSpot for marketing?”
Weak: “Is Ledgerline secure?”
Better: “Does Ledgerline support SSO, and does it have a SOC 2 report?”
Cut, then maintain
Aim for 150–300 prompts per product line. Remove near-duplicates, keep variants only where the added context changes the answer, and make sure each group has at least 20 prompts so its trend is readable. Then review the set every quarter:
Retire prompts that no longer come up in sales conversations.
Add prompts for new competitors and new buyer vocabulary.
Re-check the balance across the four intent groups.
Record every change so your trend lines stay explainable.
In our data, teams that refresh their prompt set quarterly spot new competitors in their category about six weeks earlier, measured by first appearance in Answer Diff. The set is the foundation of every number you report, so it deserves the same care as a pipeline definition.