sitenumbers.
AI search and AI crawlers

Not every AI bot visit means the same thing.

Some AI bots collect training data. Some build the index that AI search answers from. And some read your page right now, because a real person just asked an AI assistant a question. Sitenumbers tells them apart, so you know which visits actually matter.

Three kinds of AI bots

What each visit tells you.

Each AI company runs separate bots for separate jobs, and says so in its own documentation. Sitenumbers recognises each bot by name and checks that it really comes from the company it claims to.

01

Reading for a person, right now

Someone asked an AI assistant a question, and the assistant opened your page to answer it.

This is a visit on behalf of a real person. It never shows up in your visitor numbers, because the person may never click through. It is the closest thing to a direct measure of how often AI assistants use your content.

ChatGPT-UserClaude-UserPerplexity-User
02

Indexing for AI search

Your page is being added to the index an AI search answers from.

Being indexed is what makes it possible for an assistant to find and cite your page later. Pages these bots never fetch can't show up in their answers.

OAI-SearchBotPerplexityBotGooglebotBingbotApplebot
03

Collecting training data

Your page was collected as material for training AI models.

Useful to know, and something you may want to allow or block. It doesn't by itself mean your page will appear in AI answers.

GPTBotClaudeBotMeta-ExternalAgentBytespider
Read versus clicked

See how often AI reads you, and how often people click.

Other tools only see the people who click a link in an AI answer. Sitenumbers also sees how often assistants read your pages for their users, so you can compare the two, page by page.

Honest limits

What you can know, and what you can't.

AI analytics is easy to oversell. Here is exactly where the line is.

What Sitenumbers shows

  • Which pages AI assistants read for their users, and how often.
  • Read versus clicked: how many of those reads turn into visits.
  • Coverage: which pages AI search bots have indexed, and which they never fetch.
  • Blocks and errors: when bots get 403s or server errors, for example from a firewall rule nobody meant to apply to them.
  • Freshness: when each bot last read each page.
  • Real or fake: whether a bot really comes from the company its name claims.
  • People who arrive from AI assistants, and whether they convert.

What no analytics tool can know

  • What the person asked the assistant, or what it answered.
  • Whether the answer quoted or linked your page after reading it.
  • Where your page ranks inside AI answers.
  • Whether your content actually ended up in a trained model.

Some assistants also send people without a referrer, so part of their traffic looks like direct visits. Sitenumbers says so on the dashboard instead of guessing.

How it works

Counted where the bots arrive.

AI bots don't run JavaScript, so the tracking script you add for people never sees them. Sitenumbers counts them on your server or CDN instead. Pick the way that fits your site.

Only the bot's name, the page, the response status and the time are kept. Bot IP addresses are used once to confirm the bot is genuine, then discarded.

Ways to connect

Ask your AI assistant

Coming soon

Copy a ready-made prompt from your site settings into Claude Code, Cursor or ChatGPT. It adds crawler tracking to your own stack, whatever it is.

Cloudflare, one click

Coming soon

Connect your Cloudflare account and Sitenumbers sets everything up for you. No code, no deploys.

WordPress plugin

Coming soon

Install the plugin, paste your site key, done.

Next.js and Node

Coming soon

One package and one line of middleware for Next.js, Express and edge functions.

Cloudflare Worker

Ready

Deploy our Worker on your Cloudflare zone. It passes every request through untouched and reports crawler hits in the background.

nginx log shipper

Ready

A small script reads your access log every minute and sends only crawler hits. Nothing else leaves your server.