Menchly research note
High-intent prompt intelligence is not prompt volume
A practical framework for prioritising the questions closest to consequential decisions when no universal, auditable prompt-volume dataset exists.
Executive summary
Keyword volume describes aggregate search behaviour within a defined tool and methodology. Prompt intelligence examines decision-rich questions, including audience, context, criteria, alternatives and risk. The two can inform each other, but they are not interchangeable.
Keyword volume describes aggregate search behaviour within a defined tool and methodology Prompt intelligence examines decision-rich questions, including audience, context, criteria, alternatives and risk The two can inform each other, but they are not interchangeable.
Working definitions
Language leaders can use precisely.
“A high-intent prompt is a question whose context, constraints or comparison criteria indicate movement toward a consequential decision - not merely curiosity about a category.”
“Prompt intelligence is the governed practice of identifying, grouping and testing decision-relevant questions; it is not a claim to know universal prompt volume.”
Why keyword logic is an incomplete proxy
Search keywords remain useful evidence of language and demand. But conversational prompts can contain several jobs at once: education, comparison, suitability testing, objection handling and validation. They may be long, private and highly specific. A low-frequency question can still matter commercially when it precedes a high-value or high-risk decision.
Volume can also conceal mixed intent. A broad phrase such as “best luxury hotel” may represent inspiration, editorial research, a school assignment or an imminent booking. A prompt specifying travelling party, occasion, privacy needs, location and service expectation gives a clearer decision context even if no external tool assigns it a large number.
Search keywords remain useful evidence of language and demand But conversational prompts can contain several jobs at once: education, comparison, suitability testing, objection handling and validation They may be long, private and highly specific A low-frequency question can still matter commercially when it precedes a high-value or high-risk decision.
Score decision value before visibility opportunity
A practical prompt universe can be assessed across five dimensions. Audience relevance asks whether the question comes from a buyer or adviser the brand is equipped to serve. Decision proximity asks whether it supports discovery, comparison, validation or commitment. Commercial materiality considers the value or reputational consequence of the decision without inventing a forecast.
Strategic fit tests whether the brand has a truthful, differentiated reason to be considered. Evidence feasibility asks whether that reason can be supported through approved facts and credible sources. Only after these dimensions should teams consider observed model presence or competitive whitespace.
A practical prompt universe can be assessed across five dimensions Audience relevance asks whether the question comes from a buyer or adviser the brand is equipped to serve Decision proximity asks whether it supports discovery, comparison, validation or commitment Commercial materiality considers the value or reputational consequence of the decision without inventing a forecast.
Create an evidence stack, not a synthetic volume number
Use multiple inputs with explicit labels. First-party sources may include anonymised enquiry themes, approved sales notes, site search and client-adviser interviews. External sources may include search trend tools, industry forums, regulatory guidance, specialist media and competitor language. Model observations can show how assistants currently frame a question, but they do not reveal total demand.
Testing should record assistant, model where available, date, account state, retrieval mode, geography, exact prompt and output. Microsoft’s human-AI interaction guidance recommends communicating system capabilities and supporting efficient correction. Applied here, decision-makers should be shown what the observation can and cannot establish.
Use multiple inputs with explicit labels First-party sources may include anonymised enquiry themes, approved sales notes, site search and client-adviser interviews External sources may include search trend tools, industry forums, regulatory guidance, specialist media and competitor language Model observations can show how assistants currently frame a question, but they do not reveal total demand.
Important limitations
Interpret with care.
- AI assistant outputs vary by model, version, retrieval method, location, account context and prompt wording.
- The frameworks in this note support strategic judgement; they do not predict or guarantee recommendation, ranking, revenue or reputation outcomes.
- Examples are illustrative composites, not client work, performance claims or evidence that any named or implied brand has adopted this approach.
Sources
Further reading and primary references.
- Guidelines for Human-AI InteractionMicrosoft Research / CHI 2019
- AI RMF PlaybookNational Institute of Standards and Technology
- Google Trends: FAQ about Trends dataGoogle
- Search Quality Rater GuidelinesGoogle
Private AI visibility assessment
Turn the strategic question into a brand-specific baseline
Establish how assistants interpret your brand today, which questions matter, and where infrastructure can improve speed, learning and compounding results.
Request a private assessment