Buyer-Intent Prompt Pack Builder

Buyer-intent prompts are not keywords

A keyword is a compressed query aimed at a list of links: "best crm". A buyer-intent prompt is the full sentence someone types into an assistant when they want a decision made for them: "we're a 40-person B2B SaaS company outgrowing spreadsheets, which CRM should we pick if we need Slack and a real API?" The context that a search engine strips away is exactly what an AI engine uses to pick a winner.

That changes what you measure. There is no rank position to track, because there is only one answer. What matters instead is whether the answer names you, what it claims about your pricing and capabilities, which competitor it recommends first, and which sources it leans on to justify the recommendation. Constraints matter too: adding a budget, a team size or a required integration often produces a noticeably less flattering answer than a clean question does — which is why the prompts this tool generates carry your segment and integration rather than leaving them generic. Our guide to how buyers actually phrase AI questions goes deeper on wording.

How to pick which prompts to monitor

Generate broadly, then cut. A pack you re-run forever should be small enough that you genuinely read every answer. Three filters help:

• Would a real buyer type this within a month of signing a contract? If it reads like market research rather than vendor selection, drop it.
• Does the answer change who gets the deal? Comparison, alternatives and category questions move revenue. "What is a CRM?" does not.
• Can you act on a bad answer? If the fix would be a pricing page, a docs page, an integration listing or a review-site presence, keep it — that is a question worth watching.

Keep the surviving set balanced across stages rather than loading up on head-to-heads: a mix of category and alternatives questions, comparisons against your two or three real rivals, and capability, pricing and trust questions about you. Then freeze it and name it — a versioned prompt pack — because changing the questions changes the numbers and destroys your ability to compare one month to the next. The selection method is covered in full in our buyer question research guide and in which buyer questions you should monitor. If you would rather start from hand-written examples than from your own inputs, our 50-prompt library lists per-category starting packs with guidance on what to check in each answer.

How to run the pack — and what to record

Run every prompt in each engine you care about: ChatGPT, Perplexity, Gemini and Claude. They retrieve from different indexes and reach different conclusions, so the same question can name you in one and hand the deal to a rival in another. Use a fresh chat for each prompt — a running conversation contaminates the next answer — and turn off personalization or memory where you can, so you see something close to what a stranger sees. This is the manual version of prompt simulation.

For each answer, record four things: the verbatim text, which vendor is recommended first, every specific claim about you (price, tier, integration, limitation, certification) with a true or false verdict, and every source URL cited. The citations are the actionable part — they tell you which pages you would have to change, or which third-party page you would have to earn a mention on, to change the answer. A spreadsheet with one row per prompt per engine per date is enough to start; the CSV export above is laid out for exactly that.

How often to re-run it

AI answers are not stable. Models get updated, retrieval indexes refresh, a competitor publishes a comparison page, a review thread gains traction — and the recommendation flips without anyone telling you. A monthly cadence is the minimum that catches drift before it becomes a quarter of lost pipeline; weekly is right if your category is contested or you are actively fixing source pages and want to see whether the fix landed.

Always re-run the pack after a launch, a pricing change, a new certification, or a competitor's funding announcement — those are the moments answers go stale in ways that cost you deals. And re-run the identical set: same wording, same order, same engines. If doing that by hand stops being realistic, that is precisely the workflow Perciva's methodology automates.

Frequently asked questions

Keep going