Analysis
AI traffic
Two different things
| AI referral | AI crawler | |
|---|---|---|
| What it is | A real browser with a chatgpt.com referrer | GPTBot, ClaudeBot, PerplexityBot |
| Is there a human? | Yes | No — it is reading your page |
| Counted as a visitor? | Yes | No |
Most tools either conflate the two or miss both. Around 70% of AI referrals land in GA4's
"direct" bucket, because ChatGPT appends utm_source to outbound links
while Gemini and Perplexity mostly do not.
How we classify
A versioned rule table that reads three signals: utm_source, the
referrer hostname, and bare source names. The table is measured by a golden set — 41 cases,
required to score 100% on every CI run. If it goes stale, the build breaks.
Why crawlers are kept
We don't delete them, we label them. "Which of my pages is GPTBot reading?" is a feature in this product: it is the only real feedback loop for GEO/AEO work. Human metrics are always computed from rows with an empty bot label, so the two never mix.