How to Measure AI Referral Traffic from ChatGPT, Perplexity and Other Answer Engines
AI referral traffic is the set of visits that arrive from AI product domains — for example chatgpt.com or perplexity.ai — after someone clicks a citation or link inside an answer. It is one useful AEO metric. It is not the only one, because many people act on an answer without clicking.
Why clicks undercount impact
Answer engines often satisfy the question in the interface. A buyer may call you from the named list without ever hitting your site. Spearleaf's work on ChatGPT local recommendation churn is about being named, not about click-through. Measure both:
- Appearance / citation rate from repeated questions
- Referral sessions from AI hosts in analytics
- Branded search lift (optional secondary signal)
Hosts to watch (starting list)
Exact hostnames change. Start with a list you maintain quarterly:
chatgpt.com/chat.openai.comperplexity.aiclaude.aigemini.google.com/copilot.microsoft.com- Other assistants your customers mention
In GA4: Exploration or Traffic acquisition → Session source/medium, filter hostname contains those values, save the exploration. In Matomo: Referrers → Websites, or a custom segment on referrer host.
Channel grouping without lying to yourself
Some teams build an "AI assistants" channel group. That is fine for monitoring. Do not treat a quiet week as "AEO failed" — citation without click is common, and engines churn (see AEO by the numbers).
UTMs and shared links
When you control a link (newsletter, PDF, partner page), use UTMs. You usually cannot control how ChatGPT or Perplexity write their citation URLs. Expect plain referral host data for those.
Sanitize your own tests
When you or an agency re-ask engines during an audit, you can pollute referrals. Use a separate browser profile, block analytics on the QA machine, or filter known office IPs.
Sample weekly review (15 minutes)
- Open your AI hosts exploration; note sessions, landing pages, conversions if you have them
- Skim top landing pages for broken answer-first openings
- Spot-check three customer questions on two engines; log named/cited
- File one fix (listing NAP, robots rule, FAQ answer) rather than ten ideas
What "good" looks like early on
For many local businesses, early success is not a spike of chatgpt.com sessions. It is: correct facts when you are named, appearance on a few high-intent questions some of the time, and a short list of citation domains to pursue. Referral traffic tends to follow that work, not lead it.
Pair analytics with a question log
A durable AEO scoreboard looks like a table: question, engine, date, named?, position-ish note, cited URL, fact accuracy. Traffic charts alone cannot explain why ChatGPT does not recommend you.
The free basic AEO audit does not replace that log. The $49 full audit builds the first version with you.
Reporting to a non-marketer
Show three slides only: named rate on priority questions, top citation domains, and AI-host sessions. Leave the channel-group debate in the appendix. Owners decide faster with that trio.
What counts as AI referral traffic?
Sessions whose referrer hostname belongs to an AI product (for example chatgpt.com or perplexity.ai), typically from a clicked citation or link in an answer.
Why is my AI traffic near zero if friends see me named?
Many users never click. Being named can still drive calls and branded searches. Track mentions separately.
Does Google Search Console show AI Overviews clicks?
Google states that AI features are included in overall Search performance reporting. Use Search Console together with your analytics, and read Google\'s current AI features documentation for how clicks are counted.
Should I create a separate GA4 property for AEO?
Usually no. A saved exploration or channel group on your main property is enough to start.
Can AEOswitch set this up?
Yes. AI-referral tracking setup is included in Ongoing plans and can be scoped into Audit + Fixes. The $49 audit reports what we saw and what to track next.
More reading: answer-first pages · plans
