Features

Discover. Select. Convert.

One platform for the whole agent channel. See which AI agents discover and read your content, whether they pick you over rivals when a buyer asks, and which of those visits convert to signups and revenue — the traffic your web analytics can't see.

Discover

See which agents reach and read you

Before an agent can pick you or send you a buyer, it has to find and read you — and that traffic never touches your web analytics, because agents don't run JavaScript. Discovery reads the server side: which AI agents crawl your site, which pages and posts they pull, and which files they look for and can't find.

NorthwindLast 30 daysAll agentsExport

Agent reads over time

▲ 23% this month
GPTBotClaudeBotPerplexityBotGoogle-ExtendedAmazonbot
06001,2001,800FebMarAprMayJunJul
Jul 2026 · reads
GPTBot1,710
ClaudeBot1,490
PerplexityBot820
Google-Extended520
Amazonbot280

Top agents

AgentReads30d
GPTBot1,710 12%
ClaudeBot1,490 8%
PerplexityBot820 31%
Google-Extended520 4%
Amazonbot280 2%
Content reads · 30d
4,820
▲ 23%
Pages read by agents
128
▲ 9 new
Gaps agents hit
3
fixable
The Discover view · sample data

Agent traffic

who's reading you

Which AI agents crawl your site and how often — GPTBot, ClaudeBot, PerplexityBot, and the rest, named and counted.

Content reads

which pages they pull

The posts, docs, and landing pages agents actually read, so you know which content is feeding their answers.

Probe / Serve gap

the fixable leak

The difference between what AI crawlers request and what you serve them. High probe, low serve is a gap you can close.

Adoption signals

from your own stack

Private GitHub adoption tracking and log drains feed the picture from your own traffic, not just ours.

The Discovery score

Your Discovery score is the Probe/Serve gap read straight from your access logs: of the distinct paths AI crawlers actually requested, the share you served back. A 404 where a bot asked is a specific, fixable miss, and it is invisible to every tool that only checks public URLs — seeing it requires your traffic. Alongside it we show your agent readiness level from Cloudflare's scanner, rescanned weekly, with a ranked punch list of what to add.

Adoption signals from your own stack

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 — so the picture runs on your data, not just ours.

Select

Know whether agents pick you

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 pinned model versions. You win or you don't. The proportion you win is your Pick Rate, reported with a confidence interval.

NorthwindFeature flagsLast 28 daysExport

Pick Rate ranking

who agents choose
  • #1Flagport31%
  • #2NorthwindYou24%
  • #3Togglr12%
  • #4Rollout8%
  • #5Beacon3%
  • #6Switchboard1%

Your Pick Rate by model

  • Claude28%
  • GPT22%
  • Gemini21%

Weighted across surface and task, 836 trials.

Your Pick Rate
24%
▲ 3 pts
Rank
#2
of 6
Default Rate
19%
unprompted
Shortlist Rate
68%
named at all
The Select view · sample data

Pick Rate

the headline metric

How often agents choose you when they must choose. On coding tasks we parse the package actually imported; on conversational tasks a separate judge marks the primary recommendation.

Default & Shortlist Rate

the supporting cuts

Default Rate: how often agents commit to you without deliberating. Shortlist Rate: how often you're considered at all. Known-but-not-picked is a different problem than unknown.

Per-model breakdown

where you win, where you don't

Your number on each model and each surface (conversational vs coding agent), so you know whether a loss is universal or one ecosystem's.

Public leaderboards

the scoreboard

Every category has a live, public leaderboard on a rolling 28-day window. Your buyers — and their agents — can already see it.

Read the methodology → Browse the leaderboards →

Convert

Trace revenue back to the agent that drove it

Pick Rate tells you agents choose you. Agent Attribution tells you when that choice became a signup or a sale, and which agent did it. Install it on your own product and every conversion traces back to the agent that drove it. The headline is your Agent-Attributed Conversions — and the part no other tool can show is that we tie it to your Pick Rate, so selection and revenue sit in one place.

NorthwindLast 30 daysAll agentsExport
Agent-attributed conversions · 30d
342
▲ 18%
Revenue influenced
$28,400
▲ 22%

Conversions by agent

AgentConv.RevenueConfidence
ChatGPT128$11,200Confirmed
Claude96$8,400Confirmed
Perplexity61$4,900Likely
Google AI Overviews34$2,100Likely
Other agents23$1,800Inferred

By confidence

  • Confirmed224
  • Likely95
  • Inferred23

You only pay for Confirmed. Likely and Inferred are free context.

The Convert view · sample data

Every conversion is labeled by confidence

Most tools imply a precision they don't have. We tell you how sure we are about each conversion, and we lead with the ones we can prove.

Confirmed

highest certainty

The customer clicked a tagged agent link. A deterministic chain, not a guess.

Likely

strong signal

They arrived straight from an AI assistant's domain.

Inferred

lowest certainty

Self-report or timing lines up. Stated, never claimed as proof.

Influenced Conversions

the agent-era view-through

An agent read your page, handed the customer no link, and a conversion followed. The sale your analytics files under "direct" — surfaced as its own line, marked as correlation, never dressed up as a click.

Agent Reads & Top Agents

who reads, who converts

Which AI agents pull your pages mid-answer, and which ones actually convert — so you know where to invest your agent-facing surface.

Server-first. Install @pickrate/attribution and make three calls, or POST to the events API from any language; pr.js covers browser funnels. Rolling out now to teams already on our leaderboards. Read the integration guide →

Evals & alerts

Run it like a channel, not a report

Custom Evals

your matchup, your prompts

Any competitor set, your own or AI-generated prompts, the models you care about, one-off or recurring. Trace-level data in the viewer — the full prompt, the response, the scoring.

Alerts

know the moment it moves

Set a Pick Rate threshold or a rank-move trigger and get an email the moment a run crosses it. No dashboard babysitting.

Exports & share links

take the data with you

CSV/JSON export on evals and attribution, share links for reports, and webhooks that push conversions into your own stack as they happen.

Team-ready claims

one login, every tool you rep

Verified company claims gate the private data. One account can hold claims on multiple companies and switch between them in the dashboard.

Platform

Built API-first — for your stack and for agents

Everything on the dashboard is reachable by machine. That includes the agents themselves: Pickrate publishes the same discovery surfaces we audit you on.

REST API

public + authenticated

Free public endpoints for reports and leaderboards; Bearer-keyed /api/v1 for evals; secret-keyed export API for attribution.

Webhooks

push, signed

Conversion events POSTed to your endpoint as they happen, HMAC-signed (X-Pickrate-Signature), managed self-serve from settings.

SDK & browser helper

@pickrate/attribution + pr.js

A zero-dependency npm SDK for the server, a drop-in script for the browser. Publishable keys can't send revenue events.

MCP server

agents query us directly

Live Pick Rate over Model Context Protocol — getPickRate, getLeaderboard, lookupTool — plus an ARD registry and WebMCP.

Machine-readable site

llms.txt, .md, OpenAPI

Every key page negotiates to clean Markdown, the API ships an OpenAPI spec, and datasets carry schema.org markup.

Developer docs

hosted, current

Integration guides and API reference at /docs, with a machine index at /for-agents.

Which means you never have to open this dashboard

Connect Pickrate as an MCP server and your own agent answers from your live numbers — in the chat window or the terminal you already work in.

~/northwind● pickrate connected
how are agents doing on northwind.com this month?
Called pickrate · getMyAgents (ctrl+o to expand)
Last 30 days: agents read you 4,820 times across 128 pages, up 23%.
Top agents
GPTBot1,710
ClaudeBot1,490
PerplexityBot820
Google-Extended520
Amazonbot280
Most-read pages
/docs/quickstart640
/blog/how-to-ship-ai-agents512
/pricing388
/docs/api/webhooks271
/integrations244
Conversions: 38 signups and $4,200 influenced, mostly ChatGPT off /blog/how-to-ship-ai-agents.
One thing to fix: agents hit /llms.txt 86 times and got a 404. PerplexityBot is up 31% and it's the one asking most.
write an llms.txt for this repo and open a PR
Sample data · works with Claude Code, Cursor, Codex, and any MCP client

One command to wire it up: see setup.

Security & privacy

Honest measurement, careful data handling

No raw PII

hashed on arrival

End-user emails are hashed the moment they reach us — or send your own opaque user id and no email ever does. We never hold raw PII.

Key separation

pk_ can't forge revenue

Publishable browser keys can touch and identify but never convert; conversions require your server-side secret key. Revoke either anytime.

Signed egress

verifiable webhooks

Every webhook delivery is HMAC-signed so your endpoint can verify it came from us. Signing secrets are shown once and rotate on demand.

DPA & documented processors

procurement-ready

A standing DPA, a published sub-processor list, and a security disclosure policy (RFC 9116 security.txt). See the trust page.

Plans

What's in each plan

One paid plan. Free gives you the whole funnel — Discovery, reads, and full Agent Attribution with revenue — for the last 30 days. Pro ($49/mo) removes the history window and adds the diagnosis, competitor, alerting, and Custom Eval layers. Full pricing →

FeatureFreePro
Discover
Discovery — which agents crawl & read you
Content reads — which pages agents pull
Discovery audit & Probe/Serve gap
GitHub adoption tracking & log drains
Select
Public report card & all leaderboards
Pick Rate, rank, and field size
Per-model & per-surface breakdown
Default Rate & Shortlist Rate cuts
Diagnosis cockpit — why you lose, and the fixes
Competitor tracking & trend history
Convert
Agent Attribution — which agents drove each signup
Confidence tiers (Confirmed / Likely / Inferred) + revenue
Signed webhooks & CSV export into your stack
History window30 daysUnlimited
Evals & alerts
Threshold & rank-move alerts
Custom Evals (your competitors, your prompts)
Eval API (/api/v1) & CSV/JSON export

Questions

How is this different from Profound, Peec, or a GEO tool?

Tools like Profound and Peec tell you if AI mentions you: share of voice, how often you're talked about. Pickrate tells you if AI picks you. When an agent must choose you or a competitor on a real task, does it pick you? And with Agent Attribution, whether that pick turned into a signup. Mentioned isn't chosen; chosen isn't paid.

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?

The three observable stages of the agent channel: Discover (which agents crawl and read your content), Select (whether they pick you over rivals when a buyer asks — your Pick Rate), and Convert (which of those visits become signups and revenue). Pickrate measures all three in one place and regresses the discovery inputs against your Pick Rate to name the lever that moves your number.

Can I run my own evals with my own competitors and prompts?

Yes. Custom Evals (Pro) let you define any competitor set, write or AI-generate 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.

What is Agent Attribution?

It's the bottom of the Agent Funnel. Install it on your own product and it traces each signup or sale back to the AI agent that drove it. Your headline is Agent-Attributed Conversions, and the thing only we can do is tie it to your Pick Rate, so the selection score and the revenue it produces live in one place.

How is this different from GA4's AI channel or a brand-visibility tool?

Those count AI traffic or how often you're mentioned. Agent Attribution measures whether an agent's choice converted, names the agent, and labels every conversion by confidence: Confirmed (a tagged-link click), Likely (an AI-domain referral), or Inferred (self-report or timing). It also surfaces Influenced Conversions, the agent-driven sale your analytics files under 'direct.' And it connects all of it to your Pick Rate, which a generic analytics tool can't.

Do you store our users' email addresses?

No. Send an email and we hash it the moment it arrives, then store only the hash. Or send your own opaque user id and no email ever reaches us. Either way Pickrate never holds raw PII.

How do we install Agent Attribution?

It's server-first. Install @pickrate/attribution and make three calls (touch, identify, convert), or POST those events to our endpoint from any language. A browser helper (pr.js) covers client-side funnels. It's rolling out now to teams already on our leaderboards.

See the agents on your site

Free. Paste a log and see which AI agents read you in two minutes — no account needed.