How to Track AI Assistant Traffic

ChatGPT, Gemini, Claude and Perplexity now send real visitors to websites. Their bots also read your pages every day — to train models or to answer someone's question right now. You can see both. Here are three places to look, from the easiest to the most complete.

1. Google Analytics 4: the ai-assistant medium

Since July 2026 GA4 groups visits from different AI assistants under one medium: ai-assistant. You'll find them in Reports → Acquisition → User acquisition.

1. Open the User acquisition report.
2. Click Add filter.
3. Dimension: First user source / medium, match type: contains, value: ai-assistant.
4. Apply.

Now every assistant gets its own row: chatgpt.com, gemini.google.com, claude.ai, perplexity.ai, copilot.com and so on. These visits are labelled very clearly, so you see exactly who sends people to you and how many of them come back.

Which pages AI sends people to

Open Reports → Engagement → Landing page and add the same kind of filter: Session source / medium contains ai-assistant. Now you see the pages AI assistants cite and people actually click.

What GA4 doesn't show

This is clean, well-labelled traffic — but not all of it. Many assistants and apps don't pass a referrer (Grok, for example, or links copied from a chat), so those visits land in Direct. Treat the GA4 number as the minimum, not the total.

2. Cloudflare: AI Crawl Control

If your site is behind Cloudflare, open AI Crawl Control → Overview. GA4 shows people; this shows the bots that read your site.

What you can see there:

• Summary — total requests from AI crawlers, the change vs the previous period and how many got an HTTP 200.
• Metrics — total, allowed and unsuccessful requests, plus referrals. A jump in unsuccessful requests often means you block bots by accident.
• Filters — pick one crawler or several at once, or a whole operator (OpenAI, Anthropic, Meta…).
• Most crawled paths — the exact URLs bots visit most.
• AI Answer volume — for each operator: how many requests it made to build answers and how many people clicked through to your site.

Tip: compare Most crawled paths with your GA4 landing pages. Pages that bots read a lot but nobody clicks are the first ones to improve.

3. Server logs: every AI bot, no extra tools

Logs are the most complete source: every request with its user agent, on any hosting. Where to find them:

• nginx: /var/log/nginx/access.log
• Apache: /var/log/apache2/access.log or /var/log/httpd/access_log
• Shared hosting: control panel → Logs / Raw access logs

Count hits per AI bot:

grep -oiE "GPTBot|OAI-SearchBot|ChatGPT-User|ClaudeBot|Claude-SearchBot|Claude-User|PerplexityBot|Perplexity-User|Google-CloudVertexBot|Applebot|Meta-ExternalAgent|Meta-ExternalFetcher|Amazonbot|MistralAI-User|DuckAssistBot|Bytespider|CCBot|cohere-ai|AI2Bot|YouBot|Diffbot" access.log | sort | uniq -c | sort -rn

See which pages they read:

grep -iE "GPTBot|OAI-SearchBot|ChatGPT-User|ClaudeBot|Claude-User|PerplexityBot" access.log | awk ' {print $7} ' | sort | uniq -c | sort -rn | head -20

AI bot user agents to look for

Two things to keep in mind:

• Google-Extended and Applebot-Extended are robots.txt tokens, not user agents. You won't see them in logs — they only decide whether your content can be used for AI.
• A user agent is easy to fake. To be sure a hit is real, check the IP: OpenAI and Perplexity publish their IP ranges, Google and Apple can be verified with a reverse DNS lookup.

Putting it together

• GA4 — people who clicked from an AI answer, and the pages they landed on. The minimum, not the total.
• Cloudflare — which bots read which pages, and how many click-throughs each operator brings.
• Server logs — every AI bot hit, on any hosting.

Want to know how AI assistants see your site — and what to fix first? Get a free 10-minute audit.