How to Track AI Traffic in GA4: See ChatGPT, Perplexity and Gemini Referrals

GA4 tracks most AI traffic automatically. Since 13 May 2026 there is a native AI Assistants channel in the default channel group, and sessions referred by ChatGPT, Gemini, Copilot, Deepseek and Grok are tagged with the medium ai-assistant and filed there without any setup. Perplexity is not named in Google's definition and still lands in Referral, and clicks from Google AI Overviews are counted as Organic Search and cannot be separated at all. To see the full picture you need a custom channel group, which takes about ten minutes.

That is the short version. The longer version matters, because the gap between what GA4 reports and what is actually happening is where most teams either panic about a traffic drop that is not real or quietly under-count a channel that is growing fast.

Does GA4 track AI traffic automatically?

Partly. Google added AI Assistants to the default channel group in May 2026. The rule is mechanical: if the referrer matches Google's internal list of AI assistants, GA4 sets the medium to ai-assistant and the campaign to (ai-assistant), and the session falls into the AI Assistants channel.

Google's own documentation describes the channel as covering users who arrive "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok". The word doing the work there is "like". Google does not publish the complete list of domains it matches, and it does not name Perplexity, which is one of the larger referrers for a lot of B2B sites. So the channel is a good default and a bad audit.

How to see AI traffic in GA4

If you only want the built-in view, it is three clicks:

1. Open Reports, then Acquisition, then Traffic acquisition.
2. Set the primary dimension to Session default channel group.
3. Look for the AI Assistants row.

To see which assistant sent the traffic, change the dimension to Session source and add a filter, or add Session source as a secondary dimension. That is where you will notice the first problem: some AI sources you expected are not in the AI Assistants row at all.

What GA4 gets right and what it misses

Here is how the main sources are classified by default, and what each one needs from you.

The last row is the one people argue with, so it is worth being blunt about it. A click from an AI Overview happens on a Google results page and arrives with google.com as the referrer, exactly like a normal organic click. There is no parameter that reliably distinguishes it. Any tool claiming to report AI Overview clicks separately in GA4 is inferring, not measuring.

How to build a custom channel group that catches Perplexity

A custom channel group sits alongside the default one and lets you define your own rules. It applies to historical data as well, which is the main reason to bother.

1. Go to Admin, then Data display, then Channel groups, and choose Create new channel group.
2. Name it something obvious like "AI search".
3. Add a new channel called AI Assistants with the condition Session source matches regex, using a pattern such as chatgpt|openai|perplexity|claude|anthropic|gemini|copilot|deepseek|grok|you\.com.
4. Move that channel above Referral and above Organic Search in the ordering. Channel rules evaluate top down and the first match wins, so a channel sitting below Referral will never catch anything.
5. Save, then use Session custom channel group as your dimension in reports and explorations.

Step four is where most setups quietly fail. The group looks correct, the regex is right, and the row stays empty because Referral matched first.

Check your regex against reality rather than against a blog post. Open Traffic acquisition, set the dimension to Session source, and search for each assistant name to see the exact hostnames appearing in your own property. Referrer formats change, and the list above is a starting point, not a permanent answer.

Why your AI traffic numbers look too small

Almost every site under-counts this channel, for reasons that have nothing to do with your setup being wrong.

Some assistants strip or mask the referrer, so the session arrives as Direct. Desktop apps, mobile apps and API-driven integrations frequently pass no referrer at all. A person who reads an answer, does not click, and searches your brand name an hour later shows up as branded organic or direct, and no analytics package will connect those two events. And AI answers are read far more often than they are clicked, which is the entire point of the format.

The practical consequence: treat GA4's AI numbers as a floor, not a measurement. A steady rise in AI Assistants sessions plus a rise in branded search is a much better signal than either number alone. If you are stitching that against paid channels and revenue, pulling everything into a single marketing dashboard saves the weekly export ritual and makes the trend readable next to the channels you already report on.

AI search visibility metrics worth reporting

Referral sessions are one of two halves. The other half is whether you get mentioned at all, which GA4 cannot tell you, because a citation that nobody clicks leaves no trace in your analytics.

A reporting set that survives scrutiny usually looks like this:

• AI Assistants sessions and their trend, from the custom channel group above.
• Conversion rate of that channel against organic. AI referrals often convert better, because the assistant has already done the filtering. This is the number that gets budget approved.
• Prompt-level visibility, meaning how often you appear when a buyer asks a question in your category. That requires a tracker, and the tools that do it are compared in our roundup of generative engine optimization software.
• Branded search volume from Search Console, as a proxy for the reading that never became a click.

How do you measure generative engine optimization?

You measure it from both ends at once. Prompt tracking tells you whether AI engines mention your brand when someone asks a question you should own, and analytics tells you what happened to the people who clicked through. Neither is complete. Prompt tracking is a sample of a fixed question set rather than a census of real conversations, and GA4 misses every session where the referrer is stripped. Read together, they are enough to tell whether the work is paying off. Read separately, either one will mislead you.

Once the measurement is in place, the actual lever is the content. AI systems cite pages that answer a question directly, in plain language, with the answer near the top and something specific in it worth quoting. That is a different writing job from ranking a page, and it is the one covered in how to get cited by ChatGPT and in our guide to generative engine optimization.

A sensible order to do this in

Build the custom channel group first, because it is free and it applies retroactively, so you get a trend line immediately instead of starting a clock. Watch it for a month next to branded search. Only then decide whether a paid tracker will tell you something you will act on. Buying a $200 per month dashboard to confirm that you publish nothing is an expensive way to learn something you already knew.