Methodology

How we measure AI prompt demand

"Prompt volume" numbers in this industry are often browser-extension panel data — desktop-only, demographically skewed — extrapolated to precise-looking totals. We take a different approach: honest inputs, stated methods, and bands instead of false precision.

1. Measured AI search volumes, where they exist

Where our data providers publish AI search volumes for a prompt, we show them — per market where the provider supports your country, with the market stated.

2. Search-demand modeling, where they don't

AI assistants answer buyer questions by running their own web searches — we capture those query fan-outs directly from tracked answers. For prompts without measured AI volume, we anchor demand to the search volume of the underlying seed queries: if the engines search "best CRM for small business" to answer a prompt, that query's measured demand is the honest proxy. This mirrors the approach used by the most methodologically transparent tools in adjacent categories, and it uses fully licensed search data — no scraped conversations, no consent-questionable panels.

3. Bands and confidence, not fake exactness

A modeled estimate presented as "12,847 prompts/month" is a story, not a measurement. We present demand in bands, and where we compute rates from samples (like your brand's mention rate) we use proper confidence intervals — the same statistics we'd want if we were the customer.

4. Real conversations, used honestly

For topic and intent research we draw on openly licensed corpora of real AI conversations (with attribution) and our own tracked-answer corpus — for understanding what people ask and how engines respond, not for inventing absolute volume claims.

Questions about the method? Ask us — we'll show you the working. See demand data in action in the free audit.