See whether AI agents pick you — and why
Every other tool counts citations and mentions. Pickrate measures the moment that actually moves revenue: when an agent has a developer's job to do and could use you or a competitor, how often does it choose you? Then it tells you what to fix.
Pick Rate — the headline metric
Pick Rate is the share of the time an agent picks your tool over its competitors on a real, unbranded task. We ask "add transactional email," never "use Postmark," and run it many times across Claude, GPT, Gemini. On coding tasks we parse the package actually imported; on conversational tasks a separate judge marks the primary recommendation. The proportion you win is your Pick Rate — reported with a confidence interval, not a single screenshot.
The Agent Funnel — the diagnosis loop
A Pick Rate alone tells you where you stand, not why. The Agent Funnel is the four observable stages of being chosen, and it's the moat: Pickrate treats the first three as inputs regressed against the Recommendation outcome, so you learn which lever actually moves your number.
Awareness
corpus presence
Are you in the training data and the docs agents read?
Discovery
Probe / Serve gap
Can an agent's crawler actually reach and parse you?
Recommendation
the outcome
Do agents pick you when it counts? This is your Pick Rate.
Adoption
installs that stick
Once picked, do agents install you and keep you?
The diagnosis cockpit names the prompts you lose most, the rival taking them, and a real model quote as evidence — then leads with the fix: win back a specific prompt on a specific surface.
Custom Evals — run your own
The public leaderboards use our standard suites. Custom Evals let you measure exactly your matchup: any competitor set, your own prompts (or AI-generated ones), the specific models you care about, one-off or recurring. You get trace-level data — the full prompt, the model's response, and the scoring — in an in-product viewer, plus CSV/JSON export, share links, and the /api/v1 API.
Discovery — agent-readiness, audited
Before an agent can pick you, it has to be able to read you. Discovery audits your agent-readiness surface — llms.txt, OpenAPI, MCP, and robots.txt — and pairs it with the bot-log Probe/Serve gap: the difference between the AI crawlers probing you and what you actually serve them. A high Probe and a low Serve is a fixable leak.
Integrations & Alerts
Wire in the adoption signal: private GitHub adoption tracking shows whether agents that pick you actually keep you, and log drains feed the Probe/Serve gap from your own traffic. Alerts watch your Pick Rate and rank — set a threshold in points or a rank-move trigger and get an email the moment a run crosses it.
Questions
How is this different from a GEO or citation tracking tool?
GEO and brand-radar tools count mentions and citations — how often you're talked about. Pickrate measures the selection moment: when an agent has a job to do and could use you or a competitor, does it pick you? Being cited is not the same as being chosen.
Which models does Pickrate measure?
Pickrate runs real developer tasks across Claude, GPT, Gemini, on pinned model versions, and reports the proportion you win with a confidence interval.
What is the Agent Funnel?
Four observable stages of being chosen by AI: Awareness (are you in the training corpus and docs), Discovery (can an agent's crawler actually reach and parse you), Recommendation (do agents pick you — your Pick Rate), and Adoption (do they install and keep you). Pickrate regresses the inputs against the Recommendation outcome to tell you which lever moves your Pick Rate.
Can I run my own evals with my own competitors and prompts?
Yes. Custom Evals (Premium) let you define any competitor set, write or AI-generate custom prompts, pick specific models, and run one-off or recurring. You get trace-level data in the viewer plus CSV/JSON export, share links, and the /api/v1 API.
See your win rate with AI agents
Free report card, no account needed. Upgrade when you want the why.