Home / AI log analyzer
Free tool updated 2026-11-22

Which AI agents actually read your site?

An audit says what AI agents may fetch. Your access log says what they did. Drop one in to see which of 24 AI agents in our directory visited, which pages people asked assistants about, how many people clicked through from ChatGPT, Perplexity, Claude or Gemini, and which requests failed.

Drop access logs here (.log, .gz, JSON lines) or

Read in this tab; nothing is uploaded, and visitor IPs are never stored or sent.

(OpenAI, Google, Bing, Perplexity and Apple publish them; we download the public lists, your log stays here)

Optional: paste your robots.txt to flag fetches of disallowed paths

Where to find your log

From the command line

The same analyzer, for big logs, cron jobs and servers you SSH into. It runs from a copy of the repository (Node 22, no dependencies):

node bin/earncitations.mjs logs /var/log/nginx/access.log*.gz --robots robots.txt --verify
node bin/earncitations.mjs logs - --json < access.log    # stdin, machine-readable

The same count in Google Analytics

To track AI assistants as their own acquisition channel, add a channel to a custom channel group in GA4: Admin → Data display → Channel groups → Create new channel group (it starts as a copy of the default group) → Add new channel, named "AI assistants", with one condition, Source matches regex:

^((.*\.)?(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|copilot\.cloud\.microsoft|chat\.mistral\.ai|meta\.ai|chat\.deepseek\.com|grok\.com|you\.com|phind\.com|poe\.com|chat\.qwen\.ai|kimi\.com|kimi\.moonshot\.cn|duck\.ai|notebooklm\.google\.com)|chatgpt|openai|perplexity|claude|gemini|bard|copilot|mistral|mistral\.ai|le chat|lechat|deepseek|deepseek\.com|grok|you|phind)$

Then reorder the group so "AI assistants" sits above Referral: GA4 puts a session in the first channel it matches. Channel groups apply to past data too. The regex covers the 16 assistants this analyzer counts (ChatGPT, Perplexity, Claude, Gemini, Copilot, Le Chat, Meta AI, DeepSeek, Grok, You.com, Phind, Poe, Qwen Chat, Kimi, Duck.ai, NotebookLM): their sites by domain, and the utm_source names some add to cited links. It's the same list, so your analytics and your logs agree. node bin/earncitations.mjs logs --channel prints it.

Why not Google's own example? The regex on that help page (read 2026-11-22) starts with ^.*ai, so a link from any website whose domain ends in .ai counts as an AI assistant, and it has no pattern for Grok, DeepSeek, You.com or Phind. Ours names each assistant's domain instead. Visits that arrive with no referrer and no utm_source (some apps strip both) can't be attributed by any regex and land in Direct.

No server logs? Read your analytics export

If you have Google Analytics, Plausible, Fathom or Matomo but no access to server logs, export a table with one row per traffic source and read it here. It counts the visits that came from the same 16 assistants, and puts their key events (sign-ups, demo requests, purchases) and revenue next to everyone else's. A visit is a start; what matters is whether it turns into a lead.

In GA4: Reports → Acquisition → Traffic acquisition, set the table's dimension to Session source (or Session source / medium), then Share this report → Download file → Download CSV. Key events and Total revenue come along if the report shows them. For other tools, export the sources table. We read the first column named Session source, Source, Referrer, or failing those Name or Label, and the first named Sessions, Visits or Visitors (with Key events or Conversions, Revenue, and Landing page or Entry page if they're there). Column names in other languages aren't recognised yet: rename them. Numbers written 1.234,50 are read as such. The file is read in this tab; nothing is uploaded or stored.

How to read the result

Questions

Can I see when ChatGPT or Claude reads my site?

Yes, in your server or CDN access logs. Each operator's agents identify themselves in the user agent: ChatGPT-User and Claude-User when a person asks the assistant to read a page, OAI-SearchBot and Claude-SearchBot when they build their search index, GPTBot and ClaudeBot for training. This tool counts them for you.

Can I see visitors who came from ChatGPT?

Usually, yes. When someone clicks a link in an AI answer, their browser sends the assistant as the Referer (chatgpt.com, perplexity.ai, claude.ai, gemini.google.com), and ChatGPT also adds utm_source=chatgpt.com to links it cites. The analyzer counts those visits and their landing pages. Some apps and privacy settings strip the Referer, so treat the count as a floor.

Do visitors from AI answers sign up?

The log can give a first answer. For each person who clicked through from an assistant, we look for a later request from the same IP address and browser, within 30 minutes, to a page that means a step towards buying: signup, register, trial, demo, contact or sales, checkout, cart, thank-you (the CLI takes your own list with --goal). Each assistant gets "reached a goal page: N of M". It's a lead signal, not revenue. Shared networks can merge two people. A visit that converts later, on another device or through a separate app domain isn't counted. Pages that load only through JavaScript calls may never appear in the log. Addresses are matched in this tab while the file is read, and never kept.

Why do user-triggered fetches matter most?

A ChatGPT-User or Claude-User request means a person asked an assistant something that needed your page, right then. It is the closest thing to "AI search traffic" you can see in your own logs, and the pages it fetches are the pages buyers are asking about.

Are these numbers verified?

Partly. Anyone can send any user agent, so by default counts are what each request claimed. With IP checking on, requests claiming to be an agent from OpenAI, Google, Bing, Perplexity or Apple are checked against that operator's published IP ranges, and the rest are flagged as impersonators. Anthropic, Meta, Amazon, Mistral, DuckDuckGo and Common Crawl publish no machine-readable ranges we could find, so their agents stay unverified.

Why are requests marked "via proxy" or unverified?

Your log records your CDN's, load balancer's or proxy's address instead of the visitor's, so the IP check can't see who really asked. We spot this when the same addresses also carry your ordinary visitors, and count those requests as unverified rather than fake. To verify them, have your server log the client IP your CDN sets, accepted only from the CDN's own addresses: in nginx, real_ip_header with set_real_ip_from limited to your CDN's ranges; in Apache, mod_remoteip. Don't log a raw X-Forwarded-For header for this: anyone can send one with a Google address in it.

Why don't I see any AI agents?

Common reasons: the log only covers a few hours; your CDN serves cached pages without forwarding the request to your server, so origin logs miss it (use your CDN's logs); bot protection is challenging AI agents before they reach you; or the log format has no user agent (the "common" format doesn't).

How do I see AI assistant traffic in Google Analytics?

Add an "AI assistants" channel to a custom channel group: one condition, Source matches regex, with the regex in the Google Analytics section above, then move the channel above Referral. It lists the same assistants as this analyzer, so your analytics and your logs count the same visits. Visits that arrive with no referrer and no utm_source can't be attributed by any regex and show up as Direct.

Is my log uploaded?

No. The file is read in this browser tab and never sent anywhere. With IP checking on, each request's IP is compared in this tab with the operators' public lists and then discarded; visitor IPs are never stored or sent.

Seeing errors or no AI agents at all? The free audit shows what your robots.txt, pages and markup tell each agent.