Seeing ChatGPT and Perplexity Traffic Separately in GA4

The most concrete GEO question my clients ask is this: “Are ChatGPT and Perplexity actually sending us visitors, and if so, what do those visitors do?” Google Analytics 4 (GA4) can partly answer that, as long as you understand how the traffic lands in GA4 and which part of it never shows up at all.

This post focuses only on separating AI assistant traffic. General topics such as choosing KPIs, the difference between Search Console and GA4, and a monthly report template are covered in my SEO reporting guide, so I will not repeat them here. Because the GA4 interface changes often, I describe menu locations as concepts rather than exact click paths.

Short answer: since May 2026, GA4 assigns visits from the AI assistants it recognizes to an “AI Assistant” channel in the default channel group. To see ChatGPT and Perplexity separately, look at the session source dimension or create a custom channel group with a separate channel for each assistant. Clicks without a referrer appear as direct traffic, so treat the numbers as a lower bound.

How does GA4 recognize AI traffic? The referrer logic

When a user clicks a source link in a ChatGPT or Perplexity answer, the browser usually passes along the address that sent them (the referrer). GA4 records this in the session source and session medium dimensions; for example, the source might be chatgpt.com and the medium referral. If the link carries campaign parameters such as utm_source, GA4 reads the source from those parameters instead.

Some assistants have been observed adding their own utm_source value to links. That makes measurement easier, but it is not a behaviour the assistant companies are bound to keep. Do not tie your setup to a single method: build a definition that covers both referrer domains and known utm values.

What the built-in AI Assistant channel gives you, and where it falls short

In its May 13, 2026 GA4 release note, Google introduced dedicated measurement for AI assistant traffic. When the referrer matches a recognized assistant, the session medium is set to “ai-assistant”, the campaign name to “(ai-assistant)”, and the visit appears in the “AI Assistant” channel of the default channel group. The release note names ChatGPT, Gemini and Claude as examples, while the help page describing default channel definitions excludes Google's AI Overviews and AI Mode from this channel.

As of October 2026, when this post was written, Google has not published the full list of recognized assistants and does not state whether the change applies to historical data. I recommend checking in your own property, via session source, whether Perplexity lands in this channel.

Building a custom channel group for ChatGPT and Perplexity

If you want a persistent view your team can read, a custom channel group is the best option. According to Google's help documentation, standard properties can have 2 custom channel groups in addition to the default one, each with up to 50 channels. Custom channel groups can be applied to reports retroactively, and channel definitions are not case sensitive.

The most important rule is order: GA4 assigns a session to the first channel in the group whose definition it matches. That is why your AI channels must sit above Referral; otherwise the visits disappear into Referral. A practical start is to copy the default group and add these three channels at the top:

1) ChatGPT       — Session source matches regex:
   .*(chatgpt\.com|chat\.openai\.com).*

2) Perplexity    — Session source matches regex:
   .*perplexity.*

3) Other AI      — Session medium exactly matches: ai-assistant
   OR session source matches regex:
   .*(gemini\.google\.com|copilot\.microsoft\.com|claude\.ai).*

4) Default channels (Organic Search, Referral, Direct ...)

The leading and trailing .* are deliberate: GA4's help documentation says regular expressions look for a full match by default, and these wrappers catch subdomains and small variations in utm values. Put a recurring reminder in your calendar to update the expression as the list of assistants changes. If you want help setting these channels up, I can build the measurement setup with you as a GEO consultant.

A quick check without a channel group: report filters and explorations

If you first want to see whether there is any data at all, you do not need a channel group. Setting the primary dimension of the traffic acquisition report to session source / medium and filtering with the expressions above takes a few minutes. For deeper analysis, define an “AI assistants” segment in an exploration and combine it with these dimensions:

• Landing page: which of your pages do assistants cite?
• Key events: do these visits turn into form submissions, calls or sales?
• Engaged sessions: do visitors stay or leave immediately?
• Device category: is there a clear difference between desktop and mobile?
• First user source: did the assistant bring the user to the site for the first time?

The last item is an important distinction: session source shows where each visit came from, while first user source shows where the user first found the site. Look at both together to tell whether assistants bring in a new audience.

What you will never see in AI traffic data

GA4 data only shows the part of your AI visibility that turns into a click and carries a referrer. I recommend stating these limits clearly in your reports:

• Some clicks from mobile apps carry no referrer and appear as Direct traffic.
• If the user copies the link and pastes it into a browser, the source is lost.
• Clicks from Google AI Overviews and AI Mode come from Google, so they land in organic search, not the AI Assistant channel.
• Visits from users who decline cookies are under-measured, depending on your consent settings.
• Answers in which an assistant mentions your brand without a click never appear in GA4.

So read assistant traffic as a lower bound, not a precise total. I cover the impact of Google's own AI features separately in my post on Google AI Mode in Turkey.

Turning AI traffic data into GEO decisions

Separated data earns its real value when it reveals which content assistants cite. Look at what the top pages in your landing page list have in common: clear definitions, comparison tables, original data? Bring those qualities to your other important pages.

If Perplexity traffic is weak, the steps in my post on getting cited in Perplexity help; to measure clickless mentions in ChatGPT, use the question-set method from brand visibility in ChatGPT. For linking measurement to strategy, my GEO implementation guide offers a solid framework.

If you would like the channel group set up in your property or your current setup audited by an SEO consultant, you can reach me through the contact page.

8+ years of hands-on SEO and GEO experience across 150+ projects in e-commerce, SaaS, healthcare and local business sectors. Writes based on real client data, not theory.