Prompt coverage analysis: mapping the buyer questions you're absent from

Keyword research gave marketing a shared map for twenty years: a finite list of queries with volumes attached. Generative engines dissolved that map. Buyers now ask long, conversational, compound questions, like 'what should a 40-person B2B SaaS company use to monitor brand mentions in AI answers, and what does it cost?' No keyword tool will ever show you that string.

But the underlying intents are not infinite. Across a category, buyer prompts cluster into a few hundred recurring shapes. Prompt coverage analysis is the discipline of enumerating those shapes, testing them against real engines, and measuring what percentage your brand appears in. The metric that falls out, prompt coverage, is the generative-era successor to 'ranking keywords.'

Coverage is the share of a defined prompt universe where your brand appears in the answer, weighted by how it appears. We score each prompt-platform pair on a simple ladder: absent, mentioned, recommended, recommended-first. A brand can have high mention coverage and terrible recommendation coverage: engines know you exist but never suggest you, and those are very different problems with very different fixes.

Coverage is also platform-specific. It is common to see a brand cover 70% of prompts on Perplexity, which retrieves live sources, while covering 30% on Gemini, which leans harder on structured entity data. A single blended number hides that; a coverage matrix (prompts down the side, platforms across the top) exposes it immediately.

Structure the universe around the buyer journey, not around your product's vocabulary. We use five stages. Category prompts ('best AI visibility platforms'). Problem prompts ('how do I find out what ChatGPT says about my company'). Comparison prompts ('Citationly vs doing this manually'). Alternative prompts ('alternatives to <incumbent>'). And validation prompts: pricing, security, integration questions asked just before purchase.

Seed each stage from real inputs: sales call transcripts, support tickets, community threads, and the People-Also-Ask fossil record. Then expand with phrasing variants, because engines are sensitive to wording in ways rankings never were. In Citationly, the GEO Optimizer's prompt coverage module handles the expansion and dedupes intents automatically; done by hand, a rigorous universe for one product line is typically 150 to 400 prompts.

The first read is almost always the same: strong coverage on branded and comparison prompts, a cliff on problem and category prompts. That cliff is the expensive part: problem-stage prompts are where buyers who have never heard of you get their shortlist formed. If you are absent there, you are not losing deals; you are never entering them.

Look next at within-stage variance. If you cover 'best GEO tools' but not 'best AI visibility platforms for enterprise,' the gap is usually a missing page, a missing proof point, or a citation gap on a source the engine trusts for the enterprise angle. Each empty cell in the matrix is a hypothesis about missing evidence, and the citation log for that cell tells you which evidence the engine used instead.

Rank gaps by commercial weight, not alphabetically. A covered prompt that drives shortlists is worth defending; an uncovered prompt with real buying intent is worth a campaign; an uncovered prompt nobody commercially relevant asks is worth ignoring. Opportunity Finder does this triage by combining coverage data with prompt intent and competitive density, but the principle holds with a spreadsheet.

Then treat every intervention as an experiment. Ship the page, fix the aggregator profile, or restructure the answer, and re-test the exact prompt cell you targeted the following week. Coverage moves slower than rankings did, but it moves visibly, and a coverage matrix that fills in month over month is the clearest picture of GEO progress a team can put in front of leadership.