Map Your AI Visibility Across the Full Buyer Funnel

You get named for "what is CRM software" and vanish from "best CRM for a 50 person startup". One visibility score hides that entirely, so this splits your monitored set by funnel stage and tells you how much of it carries a stage label at all.

The GEO problem it solves

The split is only worth as much as the labels behind it. Each monitored prompt carries an optional stage, TOFU, MOFU or BOFU, stored on the prompt and set when it was created, generated or written by hand. It is never re-derived from the answers, so a stage is as good as whoever assigned it, and it can be absent.

That absence is the part people miss. Prompts with no stage never appear in the funnel grid, and a missing stage row means no answered prompt carries that label, not zero visibility there. Read coverage before any score, or you act on a grid describing a fraction of your monitoring.

How the workflow breaks down

1. Measure label coverage. get_keyword_triggers returns a question type breakdown next to its keyword list: one row per stage plus an "Other" row for answers whose prompt carries no stage, each with an answer count and a mention rate. That Other count is the share of monitoring the funnel view leaves out.

2. Read the grid. get_matrix with funnel granularity returns one row per stage against your monitored providers, each cell carrying score, sub-scores, citation count, position and a top competitor, each row carrying its prompt count. Stages without an answered prompt are absent, not zero.

3. Name the prompts inside one stage. get_prompt_signals takes a funnel stage filter, so you can ask it for one stage directly instead of filtering a brand-wide list afterwards. It sorts prompts into attack, defend, monitor and ignore, and returns named lists for attack and defend only; monitor and ignore come back as counts.

What a run actually returns

A real set of 155 prompts across 8 topics returned per-topic coverage from 0 to 60 percent, five of the eight at 20 percent or under, and 0 attack and 0 defend prompts on the whole portfolio.

Before any funnel label is applied, that is the real state: thin nearly everywhere. Splitting a portfolio like this by stage produces three buckets that are all weak, a true answer and a useless one. The split earns its place once coverage is uneven between stages rather than uniformly low, so check the topic spread first.

What to do with the output

Read the three results in this fixed order.

First, the coverage gate. If the Other row holds more than half the answers, stop: ranking on that grid is guesswork, and labelling prompts is the cheaper move.

Second, pick the stage. Take the lowest score among the rows returned. If it is backed by fewer than five prompts, treat it as unmeasured and take the next lowest. Break ties by prompt count.

Third, name the work. Run prompt signals filtered to that single stage. If attack and defend both come back empty while monitor and ignore carry the counts, that is a result, not a failure: no prompt at that stage clears the attack or defend test right now. Move to the next stage, or accept that the funnel lens has nothing this cycle.

What this cannot tell you is whether the weak stage is worth money. A low bottom-of-funnel score describes your monitored prompts, not lost pipeline, and there is no revenue data in these three calls.

Save the score and prompt count per stage row alongside the coverage figure. Coverage moves whenever prompts are added or relabelled, so a later comparison ignoring it reads a labelling change as a visibility change.

The Complete Prompt

I want to split my AI visibility by funnel stage and know how much of my monitoring that split actually covers. If I monitor more than one brand, list them and ask me which one before pulling any data. Use the last 30 days.

1. Run get_keyword_triggers and show the question type breakdown: one row per funnel stage plus the Other row, with answer counts and mention rates. Compute the share of answers sitting in Other. If it is above half, tell me the funnel view covers a minority of my monitoring and that labelling prompts beats reading the grid, then continue only if I say to.
2. Run get_matrix with funnel granularity. Show every stage row returned with its score, sub-scores, citations, position, top competitor and prompt count, per provider. If a stage is missing, say explicitly that no answered prompt carries that label, and do not report it as zero.
3. Rank the stage rows by score, lowest first. Skip any row backed by fewer than five prompts and tell me you skipped it and why. Break ties by prompt count. Name the stage that comes out first.
4. Run get_prompt_signals filtered to that one stage. Report the attack, defend, monitor and ignore counts, then the attack and defend prompts with their share of voice and priority. If both lists come back empty, say that no prompt at that stage currently clears the attack or defend test, and repeat this step on the next stage in the ranking. If no stage produces a list, stop and say the funnel lens has nothing actionable this cycle rather than substituting another view.

Do not describe a weak stage as lost revenue, and do not promise that work at a stage will move its score. Close with the label coverage, the stage ranking, and the named prompts if there were any.