The mapped prompt space
Two hundred or more real prompts drawn from sales calls, support tickets, community threads and assistant follow-up suggestions, written the way buyers actually phrase them.
Prompt-space research maps the questions your buyers put to assistants, as opposed to the keywords they type into a search box. Prompts are longer, conversational and full of constraints, so they cluster differently from keywords and reveal buying criteria your keyword tool never surfaces.
Typically a fixed-scope one-off, delivered in eight working days.
A keyword is a compressed query typed for a machine. A prompt is a sentence written for something that understands sentences, so buyers include budget, team size, constraints and their actual situation. That extra context is commercially valuable and completely invisible in keyword tools.
Compare the two forms of the same intent. In a search box a buyer types 'crm for small business'. To an assistant the same buyer writes 'we're a 12-person agency using spreadsheets, need something our non-technical account managers will actually adopt, ideally under a fortnight to set up'. The second contains the entire qualification conversation your sales team usually has on a first call.
That is the real prize here. Prompt research does not just tell you what to write — it tells you which objections, constraints and disqualifiers your buyers volunteer before they ever speak to you. Several clients have taken prompt clusters straight into sales enablement because the language is more honest than anything a survey produces.
Prompts also cluster differently. Keyword research groups by shared terms; prompts group by shared situation. Two prompts with almost no vocabulary in common frequently belong to the same cluster because they describe the same buyer in the same bind, and a tool matching on strings will never put them together.
The sources are different too. Keyword volume comes from search data you can buy. Prompt-space research comes from your sales calls, support tickets, community threads and the follow-up questions assistants themselves suggest. There is no volume figure, which is uncomfortable for anyone used to a keyword tool, and no substitute for it either.
You receive a mapped prompt space for your category, clustered by buyer situation rather than shared vocabulary, scored for commercial intent, with the constraints and objections buyers volunteer surfaced separately. The panel becomes the measurement baseline for everything else in the pillar.
Two hundred or more real prompts drawn from sales calls, support tickets, community threads and assistant follow-up suggestions, written the way buyers actually phrase them.
Prompts grouped by the buyer's circumstance rather than by shared terms, because two prompts describing the same bind belong together even with no vocabulary in common.
Which clusters sit closest to a buying decision, so content effort goes where the pipeline is rather than where the volume looks largest.
The disqualifiers buyers volunteer unprompted — budget shape, team capability, timeline, integration worries. Frequently the most useful page for your sales team.
Which clusters you currently have no content addressing at all, ranked by how close they sit to a decision.
The prompt set formatted as a fixed panel, so every other GEO engagement measures against the same baseline rather than inventing its own.
The research runs over eight working days: harvest real buyer language from your own conversations, expand it through assistant follow-up suggestions, cluster by situation, score for commercial intent, then hand over a panel that doubles as the measurement baseline for later work.
Sales call recordings, support tickets, sales emails and community threads. We take the phrasing buyers use with humans, because it is closest to the phrasing they use with assistants.
→ Raw prompt corpus in buyers' own words
Each seed prompt run through the assistants themselves, capturing the follow-up questions they suggest. This surfaces the adjacent questions buyers ask next and would not have thought to tell you about.
→ Expanded corpus with follow-up chains
Grouping by the buyer's circumstance rather than shared vocabulary, then naming each cluster in language your team recognises rather than in tool output.
→ Named situation clusters
Commercial intent scored per cluster, gaps identified, and the whole set formatted as a fixed measurement panel. Walked through live with both marketing and sales.
→ Scored map, gap list and reusable panel
They answer adjacent questions and neither replaces the other. Keyword research tells you what people type and how many of them do it. Prompt research tells you what people ask and what constraints they bring, with no volume figure available and much richer commercial context.
| Keyword research | Prompt research | |
|---|---|---|
| Input form | Two to four compressed terms | A full sentence with context and constraints |
| Volume data | Available and reliable | None — this is the main trade-off |
| Grouping logic | Shared terms and search intent | Shared buyer situation |
| Reveals constraints | Rarely | Routinely — budget, timeline, team capability |
| Source | Search data you can buy | Your own calls, tickets and community threads |
| Feeds | Page targeting and content briefs | The measurement panel and answer structure |
You need this when you are about to invest in GEO and have no idea what your buyers actually ask, when your content answers questions nobody poses in those words, or when sales and marketing describe the same buyer using two entirely different vocabularies.
The most common finding is a gap between the vocabulary your content uses and the language buyers actually bring. The second is a cluster of high-intent situations nobody has written for, usually because they surface constraints the marketing team would rather not address directly.
Each answer is written to stand alone in 40 to 60 words — the shape an AI Overview or Perplexity citation lifts. Ships with FAQPage schema.
Keywords are compressed queries typed for a machine; prompts are sentences written for something that understands sentences. Prompts therefore carry budget, timeline and capability constraints that keyword tools never surface, at the cost of having no reliable volume data attached.
Your own sales calls, support tickets and community threads first, because that is the closest available record of how buyers phrase things to another party. We then expand each seed through the assistants themselves, capturing the follow-up questions they suggest.
No, and nobody credibly can — the engines do not publish it. We score commercial intent and cluster frequency within your own corpus instead. Anyone selling you prompt volume figures is presenting modelled estimates as measurement.
Two hundred is the standard panel and is enough to trend reliably once you start measuring. Broader catalogues justify more. What matters more than count is that the set stays fixed afterwards, otherwise later measurements are not comparable.
Frequently, yes. Several clients have taken the objection and constraint map straight into sales enablement, because buyers volunteer disqualifiers to an assistant more honestly than they do on a first call with a vendor.
They pair naturally and the audit includes prompt design. Buy this separately when you want the research depth for content and sales use beyond measurement, or when your category is unusual enough that the panel deserves its own engagement.
Thirty minutes with a senior strategist. We pull your live visibility while we talk and tell you plainly whether a prompt research is what you need — or whether your problem sits somewhere else.