The Agent Funnel: The Four Observable Stages of Being Chosen by AI
Every framework for "being chosen by AI" is either hand-wavy or unmeasurable. The Agent Funnel is neither. It's four stages, each observable, each with a metric, and the whole point is to find which one is quietly capping your Pick Rate.
Stage 1 — Awareness
Is the model aware you exist? This is corpus presence: training data, docs, the GitHub and Stack Overflow footprint models learn from. Measured by Shortlist Rate — how often you're named at all. If you're rarely shortlisted, this is your ceiling. It's the slowest stage to move, but everything downstream is zero without it.
Stage 2 — Discovery
Can an agent read you, right now, well enough to commit? This is the machine-readable surface — llms.txt, OpenAPI, MCP, robots.txt — and the Probe/Serve gap between what crawlers request and what you return. It's the most fixable stage and the most neglected. Known-but-not-picked usually fails here.
Stage 3 — Recommendation
Does the agent actually pick you on an unbranded task? This is the outcome — your Pick Rate — and it's what the first two stages exist to move. It varies wildly by category: a winner-take-most market like payments has one tool near 78%, while contested markets like email or auth are decided by a few points. Knowing which kind of category you're in changes your whole strategy.
Stage 4 — Adoption
Once picked, does it stick? Does the agent install you, keep you, and reach for you again — or rip you out on the next refactor? Adoption is the durability signal, measurable through installs and repeat selection. A high Pick Rate with weak Adoption means you win the first impression and lose the relationship.
Why it's a funnel, not a checklist
The stages compound. Perfect Discovery can't save you if Awareness is zero; a great Pick Rate leaks away if Adoption fails. The leverage is in treating the first three as inputs you can move and regressing them against the Recommendation outcome. That regression tells you which lever actually shifts your number — instead of guessing.
That diagnosis loop is what Pickrate runs for you: measure the outcome, attribute it to a stage, fix that stage, re-measure. See the live numbers and find your capping stage.
FAQ
What are the four stages of the Agent Funnel?
Awareness (corpus presence), Discovery (machine-readability and the Probe/Serve gap), Recommendation (your Pick Rate on unbranded tasks), and Adoption (whether the pick sticks). The first three are inputs you regress against the Recommendation outcome.
Which stage should I fix first?
The one that's actually capping you, which is rarely the one you assume. Measure your Pick Rate and Shortlist Rate first. A high Shortlist with a low Pick points at Recommendation or Discovery; a low Shortlist points at Awareness.
See your Pick Rate
Check how often AI agents pick your tool — free, no account needed.
Check your tool →Keep reading
We Measure Whether Agents Pick Tools. Now They Can Pick Pickrate.
Pickrate is now WebMCP-native: open it in Chrome and an in-browser AI agent can call its tools directly. Why being callable, not just recommended, is the next stage of Agent SEO.
Your Coding Agent Can Now Ask Pickrate Which Tool to Pick
Pickrate is now a remote MCP server. Add one line to Cursor or Claude Desktop and your agent can query any tool's Pick Rate, mid-task. Why being callable is the whole point.