AI Agent Analytics

Google Analytics for AI Agents

See which AI agents read your site, what they come for, which of them are fake, and how many people they send back. Even the ones that never run JavaScript.

Start Free Trial See All 23 Agents
  • Verified against published IP lists
  • Works without JavaScript
  • Agent IP addresses never stored
AI Agents report showing 29,396 agent requests from 14 agents, split by training, AI search and user-triggered fetches

Example data from a demo store.

Why It Matters Now

AI Is Now One of Your Biggest Readers

AI companies read the web at enormous scale, and their assistants are becoming a real source of visitors. Here is what independent reports measure.

53%

of all web traffic in 2025 was automated, up from 51% the year before.

Source: Imperva Bad Bot Report 2026

+82%

growth in AI agent traffic, whose share of all web requests rose from 1.2% to 1.7%.

Source: DataDome, 2026

2×

ChatGPT referral traffic more than doubled from January to August 2026.

Source: BrightEdge, September 2026

100s:1

AI crawlers request hundreds, sometimes thousands, of pages for every visitor they send back.

Source: Cloudflare Radar, 2026

Third-party figures; each report measures differently. Last checked October 2026.

Three Kinds of AI Visitors

Not All AI Agents Want the Same Thing

Every request is classified by why the agent came, because each kind deserves a different decision.

Training

Collecting Training Data

Crawlers that copy your content to train future AI models. They bring no visitors directly.

Examples: GPTBot, ClaudeBot, anthropic-ai, Meta-ExternalAgent

Your call: Allow them if you want your brand and expertise in tomorrow’s models, or block them in robots.txt.

AI search

Indexing for AI Search

Crawlers that index your pages so AI search can cite and link to them, much like a search engine.

Examples: OAI-SearchBot, Claude-SearchBot, PerplexityBot, Amazonbot

Your call: Usually worth allowing: blocking them can remove your pages from AI answers.

User-triggered

Fetching for a Person

Fetchers that load a page right now because someone asked an AI assistant about it.

Examples: ChatGPT-User, Claude-User, Claude-Web, Perplexity-User

Your call: Each one is a real person’s question. Make sure these pages load fast and answer well.

How It Works

Seeing What a Tracking Script Can’t

Most AI agents download your HTML and leave without running a line of JavaScript. Two collection methods and one check close that gap.

The Tracking Code

The same one-line tracking code you use for visitor analytics also records the AI agents that do render your pages.

The Server-Side Collector

A snippet for Cloudflare Workers, Next.js, WordPress or any server reports every agent request after your page is sent, so nothing slows down.

Daily Verification

Anyone can claim to be GPTBot. Requests are checked against the IP lists published by OpenAI, Anthropic, Perplexity, refreshed daily. Impostors are listed separately and never counted.

The Reports

Answers You Can Act On

Every Agent, Verified

One table with every agent that visited, who runs it, why it came and how much of its traffic is verified.

  • Grouped into training, AI search and user-triggered fetches
  • Verification rate for every agent that can be checked
  • Requests, pages and last visit at a glance
Agents table with GPTBot, Meta-ExternalAgent, ClaudeBot, OAI-SearchBot and PerplexityBot, their purpose, requests, pages and verification rate

Example data from a demo store.

From Crawl to Visit

The pages agents read most, next to the people who arrived on those pages from ChatGPT, Claude or Perplexity.

  • Most requested pages for every agent combined
  • Filter by a single agent to see what it reads
  • AI assistant visits for each page, to see what gets cited
Most requested pages by AI agents with the number of agents and the visits each page received from AI assistants

Example data from a demo store.

What Agents Can’t Find

Requests that end in an error show exactly what agents look for and miss.

  • Missing and broken pages, with their status codes
  • Spot agents asking for an llms.txt you don’t have yet
  • Fix the gaps and watch the errors disappear
Pages AI agents could not get, led by /llms.txt and /api/products.json returning 404

Example data from a demo store.

Agent Registry

23 AI Agents Recognized Today

Agents from 14 companies, 8 of which can be verified against IP lists their operators publish.

23 AI Agents Recognized Today
Agent Operator Purpose User agent token Verified by IP
OAI-SearchBot OpenAI AI search OAI-SearchBot Yes
ChatGPT-User OpenAI User-triggered ChatGPT-User Yes
GPTBot OpenAI Training GPTBot Yes
Claude-SearchBot Anthropic AI search Claude-SearchBot Yes
Claude-User Anthropic User-triggered Claude-User Yes
ClaudeBot Anthropic Training ClaudeBot Yes
Claude-Web Anthropic User-triggered Claude-Web No public list
anthropic-ai Anthropic Training anthropic-ai No public list
Perplexity-User Perplexity User-triggered Perplexity-User Yes
PerplexityBot Perplexity AI search PerplexityBot Yes
Google-CloudVertexBot Google User-triggered Google-CloudVertexBot No public list
Meta-ExternalFetcher Meta User-triggered meta-externalfetcher No public list
Meta-ExternalAgent Meta Training meta-externalagent No public list
FacebookBot Meta Training FacebookBot No public list
MistralAI-User Mistral User-triggered MistralAI-User No public list
DuckAssistBot DuckDuckGo User-triggered DuckAssistBot No public list
Amazonbot Amazon AI search Amazonbot No public list
YouBot You.com AI search YouBot No public list
Bytespider ByteDance Training Bytespider No public list
CCBot Common Crawl Training CCBot No public list
cohere-ai Cohere Training cohere-ai No public list
AI2Bot Ai2 Training AI2Bot No public list
Diffbot Diffbot Training Diffbot No public list

This list comes straight from the product and grows as new agents appear.

Take Control

Decide Who Reads Your Site

Turn what agents actually do on your site into rules, files and alerts.

  • robots.txt rulesChoose which kinds of agents to allow and copy ready-made rules, with the agents that visited you listed first.
  • An llms.txt draftBuilt from your public pages that people and agents use most, ready for you to review and publish.
  • AlertsAn email and a notification when a new agent arrives, an agent crawls far more than usual, or fake agents surge.
  • Email reportsAI agent activity in your weekly or monthly summary.

Generated robots.txt, blocking AI training only:

# AI agents, generated by Metrics Hive

# OpenAI
User-agent: GPTBot
Disallow: /

# Anthropic
User-agent: ClaudeBot
Disallow: /

# Anthropic
User-agent: anthropic-ai
Disallow: /

# Meta
User-agent: meta-externalagent
Disallow: /
…
Ask Your AI Assistant

Ask Claude About Your Traffic

Connect Claude, Claude Code, Cursor or any MCP client, then ask questions in plain language. Answers come from your live data.

  • Paste one URL, sign in and click Allow. No API key to copy.
  • Read-only access through OAuth, which you can revoke at any time.
  • Agent analytics, AI assistant visits and robots.txt suggestions as ready-made tools.
Which AI agents crawled my site most this month, and what do they fail to find?
get_agent_analytics
GPTBot made 7,675 requests, followed by Meta-ExternalAgent (5,840) and ClaudeBot (5,165). Agents asked for /llms.txt 1,401 times and got a 404, so publishing one is your quickest win. ChatGPT also sent you 196 visitors this month.

Example data from a demo store.

Setup

Copy, Paste, Done

Each website gets its own collector key and ready-to-use snippets, including a WordPress plugin that downloads already configured.

  • Cloudflare Workers
  • Next.js
  • WordPress
  • cURL

Simplified Cloudflare Worker:

export default {
    async fetch(request, env, ctx) {
        const response = await fetch(request);
        const user_agent = request.headers.get('user-agent') || '';

        if(AI_AGENTS.test(user_agent)) {
            ctx.waitUntil(fetch(METRICSHIVE_URL, {
                method: 'POST',
                headers: {Authorization: `Bearer ${env.METRICSHIVE_AGENTS_KEY}`},
                body: JSON.stringify({
                    user_agent,
                    path: new URL(request.url).pathname,
                    status_code: response.status,
                    ip: request.headers.get('cf-connecting-ip')
                })
            }));
        }

        return response;
    }
};
FAQ

AI Agent Analytics, Explained

What counts as an AI agent?
Software run by AI companies that reads your pages: crawlers that collect training data, crawlers that index pages for AI search, and fetchers that load a page because a person asked an AI assistant about it.
Why doesn’t my current analytics show AI agents?
Analytics tags run in a browser. Most AI crawlers download your HTML without running JavaScript, so the tag never loads, and many tools filter out bots on purpose. The server-side collector sees these requests on your server or at your edge instead.
How are fake agents detected?
OpenAI, Anthropic and Perplexity publish the IP addresses their agents use. Every request claiming to be one of their agents is checked against those lists, refreshed daily. If a list can’t be loaded, requests are marked as not checked instead of fake.
What do you store about each request?
The agent, the page path without query strings, the status code and whether it was verified. The IP address is only used for the check and is never stored.
Should I block AI crawlers?
It depends on what you want. Many sites block training crawlers but allow AI search crawlers, because those can cite and link to them. Agents fetching a page for a person may not follow robots.txt at all. Your AI Agents report shows which agents visit, so you can decide with data.
What is llms.txt?
A proposed plain-text file at the root of your site that points AI tools to your most useful pages. Your report shows whether agents ask for one, and a draft is generated from your most used public pages.
Can I ask Claude about my AI traffic?
Yes. Add the MCP server URL in Claude, Claude Code, Cursor or another MCP client, sign in and approve read-only access, then ask questions in plain language.
Is AI agent analytics included in every plan?
Each plan includes a monthly number of AI agent requests per website, listed on the pricing page.

See Every Visitor, Human and AI

Try any plan free for 7 days and keep your current analytics running alongside while you compare.

Start Free Trial