When ChatGPT, Claude, Perplexity, or Gemini cite or link your content, a real human can click that citation. The resulting visit shows up in your analytics with a referrer that names the AI surface — but only if your analytics tool understands what to do with it.
WebmasterID classifies these visits as traffic_category = ai_referral and tags the referrer source explicitly (chatgpt, claude, perplexity, gemini, and others). They stay on the human-events track because they are humans — they just arrived from an AI surface.
AI referrals vs. AI crawlers
This is a small but important distinction:
- AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, …) are bots reading your site. They are written to
bot_visits. - AI referrals are humans who clicked a link from inside an AI assistant's interface. They are written to the events table with
traffic_category = ai_referral.
Both are "AI signal", but they answer different questions — one is "who is reading me" and the other is "who is sending me readers."
What you can do with the signal
AI referral data is most useful when treated as a distribution channel rather than a vanity metric. If a particular surface is consistently sending readers, your structured-data and machine-readability investments for that surface are paying off. If a surface is invisible to your analytics, that is also information — particularly for publishers whose work is being summarised inside AI surfaces without driving back any visits.
Limitations to be honest about
AI assistants do not always preserve referrer headers. Some surfaces wrap outbound clicks in their own tracking domain (which WebmasterID still recognises in the referrer-source classifier), some send the visitor through an intermediary page, and some strip the referrer entirely. The classifier is a best-effort signal, not a perfect one.
For more on AI visibility as a whole, see /ai-visibility. For where this data lives in the dashboard, the AI Visibility page is the canonical surface.