How to measure AI traffic in GA4 (ChatGPT, Perplexity, Gemini)
TL;DR
- AI assistants send referral traffic from domains like chatgpt.com, perplexity.ai and gemini.google.com - but GA4 lumps them into Referral or Organic by default.
- Build a custom channel group with a regex to isolate 'AI Assistants' and see the channel clearly.
- Expect undercounting: some AI apps strip the referrer, so those visits land in Direct.
- Judge this traffic by engagement and conversions, not volume - it's usually small but high-intent.
A growing slice of your visitors arrive because an AI assistant cited or recommended you. By default, GA4 hides them - scattered across Referral, Organic Search and Direct. If you want to know whether your GEO work is paying off, you have to isolate that traffic deliberately.
Where AI traffic comes from
The referrers to watch are the assistant domains themselves: chatgpt.com, perplexity.ai, gemini.google.com, and Copilot (often surfacing via bing.com). When the referrer is preserved, these show up in GA4 as Referral. The problem is they're mixed in with every other referring site.
How to set it up
The clean solution is a custom channel group in GA4, with a new channel that catches these domains.
- In GA4, go to Admin, then Data settings, then Channel groups, and create a custom group (copy the default first).
- Add a new channel named 'AI Assistants', placed above Referral and Organic so it wins the match.
- Set the condition: Source matches regex, using a pattern for the AI domains.
- Apply and use the group in your reports and explorations.
- Optionally, mirror it as a segment for deeper analysis.
A regex that catches the main players:
chatgpt\.com|chat\.openai\.com|openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|bing\.com/chat
Keep it updated as new assistants appear - this list will grow.
The caveats nobody mentions
Two honest warnings. First, undercounting is guaranteed: several AI apps (especially in-app or mobile) don't pass a referrer, so those genuine AI visits fall into Direct and you'll never fully attribute them. Second, volume will look small - and that's fine. AI referral traffic tends to be low in number but high in intent: people arrive already half-convinced because a model vouched for you. Measure it by engagement rate and conversions, not by chasing a big number.
Key data
| Share of AI sessions | ~1% |
|---|---|
| ChatGPT within AI pool | ~87% |
| ChatGPT sessions in (not set) | ~22% |
| Perplexity sessions in (not set) | ~32% |
Agency-study data, not a Google source. GA4 under-attributes AI to Direct/(not set).
Sources: Workshop Digital
Common mistakes and how to do it right
Measuring AI traffic is easy to fake and easy to over-read. Keep it honest:
| Avoid | Do |
|---|---|
| ✗ Assuming GA4 shows AI traffic out of the box | ✓ Building a custom channel group with a regex to isolate it |
| ✗ Trusting the number as complete | ✓ Accepting that referrer-less AI visits leak into Direct |
| ✗ Judging AI traffic by volume | ✓ Judging it by engagement and conversion - it's small but high-intent |
| ✗ Setting the regex once and forgetting | ✓ Updating the pattern as new assistants emerge |
Tools I use and recommend
- Google Analytics 4Behaviour and conversion measurement.
- Google Search ConsoleOrganic performance, coverage and field Core Web Vitals.
- PerplexityAnswer engine with sources, useful for research and GEO.
- ChatGPTAI assistant for research, analysis and quick audits.
Sources
Frequently asked questions
Why is my AI traffic in Direct, not Referral?
Some AI apps don't send a referrer, so GA4 can't attribute the source and defaults to Direct. It's a known limitation - expect partial data. Analytics service →
Is AI referral traffic worth tracking if it's small?
Yes. It's often low volume but high intent, and it's the clearest proof your GEO work is landing. GEO guide →