AI agent analyticsMeet the AI reading your site.

Search crawlers, AI crawlers and AI agents read your pages every day. Little Friend counts them apart from people, shows how sure it is about each one, and lets you choose which get in.

Sign up freeRead the Agents docs

Every visitor in its own lane.

Automated requests have their own count. A crawler's requests never add to your page views, and each AI request counts under one purpose: AI training, AI search indexing, Assistant fetching for a user, AI agent or agentic browser.

ClassMeaning
Search crawlerIndexes pages for a search engine, like Googlebot or Bingbot.
AI crawlerCollects pages for AI training or retrieval, like GPTBot or ClaudeBot.
User-directed AI agentFetches a page because a person asked an AI assistant to, like ChatGPT-User.
Uptime and monitoringUptime checks and monitoring services.
Link previewBuilds a link preview when someone pastes your URL into a chat or a post.
Suspected automationScripted traffic that hides or misstates who runs it.
UnknownNot enough signal to say.

Names you can check.

A user agent is a claim. Every verdict carries the evidence behind it, so "GPTBot" from OpenAI's published addresses reads differently from "GPTBot" anywhere else.

EvidenceWhat it means
No evidenceNothing backs up who runs it.
Self-declared (user agent)The user agent names an operator. Anyone can send any user agent, so this is a claim.
Verified network (DNS or published IP range)The request came from the operator's network: a reverse DNS check or a published IP range matched.
Verified request signatureThe request carried a valid cryptographic signature from the operator (Web Bot Auth).
Authenticated credentialAn authenticated credential identified the client. That proves the client, not its operator.

See what each AI takes and brings back.

Visits from AI assistants

People who arrive from ChatGPT, Claude, Perplexity, Gemini, Copilot and others land in an AI assistants channel, by referrer or by the utm_source the assistant adds to its links.

Channel rules

Requests per visit

For each company that runs an assistant, Little Friend divides the AI requests it made by the visits its assistants sent back, so you see who reads a lot and sends little.

How we count

robots.txt, checked

Little Friend reads your robots.txt and lists the disallowed routes each crawler requested anyway.

Agents report

Choose which agents get in.

The door runs in your Worker, proxy or server and decides on each request before your app runs: allow, slow down, block, require a signature. Every project starts in dry run, so you see what your rules would do before they change anything.

PresetWhat it does
OpenLet every agent in and record what it does. Every project starts here.
No trainingBlock crawlers that collect pages to train AI models. Search engines and assistants still get in.
Verified onlyBlock agents that fail their operator check, and slow down agents that cannot show who runs them.
CommerceCheckout, cart and account pages need an agent acting for a signed-in user. Browsing stays open.
  • People always get in. The door only answers automated clients. Every browser is let through.
  • Signed agents, verified. Agents can prove who they are with Web Bot Auth request signatures.
  • Who they act for stays with you. The door sends Little Friend its decision, and nothing about the person behind the agent.

Set up the door How agents sign their requests

Three ways to see them.

Most crawlers never run JavaScript, so a browser script alone can't see them. Little Friend reads the requests themselves.

  • A Vercel log drain. Every request your site serves, with no code change. It needs a Vercel Pro or Enterprise team. Add a drain
  • @littlefriend/edge. For Cloudflare Workers, Express and Node, Bun, Deno and Next.js route handlers. Add the package
  • Any language. Post what your edge saw to one endpoint with an edge key. Send observations

Good to know.

  • AI citations aren't tracked. GPTBot reading a page doesn't mean an AI answer used it.
  • Agents that pass as browsers look like people. An agent that sends a browser's user agent from ordinary addresses counts as a person.
  • Unknown stays unknown. Little Friend never infers identity from an IP address and device traits.

Ask Claude about your traffic Google Analytics alternative Little Friend vs Google Analytics