How to see how much of your traffic comes from AI

The question comes up in almost every founder call now: "How much of our traffic is coming from ChatGPT?" It's the right question. The frustrating part is that if you open GA4 and go looking for the answer, you won't find it, at least not sitting in one place with a label on it.

That's not because the traffic isn't there. It's because GA4, out of the box, breaks AI traffic into pieces and files those pieces in three different drawers. The fastest-growing source of high-intent discovery a lot of brands have is hiding in plain sight, spread thin enough that nobody notices it's growing.

This guide fixes that. It assumes you already have GA4 installed and working (if not, start with the foundations guide first). We'll build one custom channel group that pulls AI referrals into a single, countable source, apply it retroactively so you can see the trend, and then I'll be honest with you about the traffic no analytics tool can catch.

A quick note on terms first. AI search analytics is the umbrella for measuring your brand across AI search, and it has two halves. One half is traffic: the clicks AI tools send to your site, which is what this guide sets up in GA4. The other half is visibility: whether AI answers name or cite you at all, click or not. This guide covers the traffic half in full. For the visibility half, start with our companion guide on how to track brand mentions in AI search. Together, the two halves are the whole picture.

Why GA4 hides AI traffic by default

When someone clicks through to your site from an AI tool, GA4 has to decide which channel that visit belongs to. The problem is that AI traffic doesn't behave like one thing, so it gets sorted into whatever bucket its technical fingerprint matches. Here's where it actually lands:

The same underlying source, AI, split across three channels. This is why "how much comes from AI?" has no default answer in GA4.

That middle row is the painful one. Industry estimates put the share of AI referral sessions that arrive with no referrer header, and therefore fall into Direct, at somewhere between 35% and 70%. A large chunk of your AI traffic is effectively invisible before you even start, quietly inflating "Direct" (the channel everyone already ignores as "people who typed the URL").

Doesn't GA4 have a native AI channel now?

It does, and it helps a little, but it is not the answer on its own. On May 13, 2026, Google added a default channel called AI Assistant to GA4, which automatically classifies traffic from a handful of AI tools using a new medium value, ai-assistant. It recognises the big names it knows about: ChatGPT, Gemini, DeepSeek, Copilot, and Grok. Useful. But read the catches before you rely on it:

• It excludes Perplexity. One of the most-used answer engines simply isn't in the native list, so it stays in Referral.
• It isn't retroactive. The native channel only classifies traffic from its launch date forward. All your AI traffic from before May 2026 stays filed wherever it originally landed, so you can't see the trend that got you here.
• It still misses referrer-less traffic. The Direct problem above doesn't go away. Clicks with no referrer header still land in Direct.
• Google's own AI still counts as Organic Search. AI Overviews and AI Mode aren't captured here.

So the native channel is a convenience, not a measurement strategy. To see AI as one source, including Perplexity, and to see it across your whole history, you build a custom channel group. That's the part that actually answers the founder's question.

Build the custom AI channel group

This takes about ten minutes and you only do it once. You'll need Editor access to the GA4 property. Here's the sequence:

1. In GA4, click Admin (the gear icon, bottom left). Under Data display, open Channel groups.
2. Click Create new channel group. Give it a clear name like AI-aware channels and a short description.
3. Click Add new channel. Name it AI Assistants.
4. Set the condition to Source  matches regex, and paste the pattern below.
5. Reorder the new AI Assistants channel so it sits above Referral and Organic Search. Order matters: GA4 assigns each session to the first channel it matches, top-down. If AI Assistants sits below Referral, Perplexity will get caught as Referral first and never reach your AI channel.
6. Click Save.

Here is a source-based regex covering the major AI engines. It's deliberately broad so new subdomains still match, and you can trim or extend it:

A note on Google AI Overviews

Notice the regex above does not try to pull Google's AI Overviews out of Organic Search. That's deliberate. Those clicks genuinely originate on a Google search results page and share the same referrer as ordinary organic clicks, so there's no clean, reliable signal to separate them in GA4 without guesswork. Trying to force it usually creates more measurement error than it removes. For the AI Overviews picture, lean on Search Console and a dedicated visibility check instead, which I'll come back to.

How to actually read it

Creating the group doesn't change your standard reports automatically. You choose to view your data through it. Two ways:

• In the Traffic acquisition report: go to Reports → Acquisition → Traffic acquisition. Click the dropdown on the primary dimension (it defaults to "Session default channel group") and switch it to your new AI-aware channels group. Your AI Assistants channel now appears as its own line, with sessions, engagement, and conversions beside it.
• In an Exploration: go to Explore, start a blank report, and add your custom channel group as a dimension. This is where you can break AI traffic down by landing page (which content AI sends people to), by conversion, and over time.

Once it's in front of you, the questions worth asking are: what share of total sessions is AI now, and which direction is it moving? Which pages are AI engines actually sending people to? And, most importantly, how does AI traffic convert compared to your other channels? In a lot of the accounts I look at, AI referrals are still small in volume but convert well, because the person arrived with intent already formed by the answer they read. Small and high-intent is exactly the kind of channel you want to catch early.

The honest limits: what GA4 still can't tell you

This is the part most tutorials leave out, and it's the part that matters most. Your custom channel group makes AI traffic visible and countable, which is a real step up. But it measures one thing only: clicks that reached your site. Two big blind spots remain, and no amount of GA4 configuration closes them.

1. The referrer-less Direct problem doesn't fully disappear. A meaningful share of AI clicks arrive with no referrer at all and land in Direct, indistinguishable from someone typing your URL. You can narrow this gap where you control the links, by adding UTM tags to any URLs you place inside content that AI tools ingest, but you cannot fix the clicks that come through channels you don't control. Treat your AI number as a confident floor, not a precise total.

2. GA4 can't see the answers where nobody clicks. This is the bigger one. AI answer engines increasingly resolve a user's question inside the answer itself. Your brand can be named, quoted, and recommended to a high-intent buyer who never clicks through, because they already got what they needed. That is a genuine visibility win, and it produces exactly zero sessions in GA4. Web analytics can only ever measure the traffic that arrived, never the influence that didn't need a click.

The fuller picture: pair three views

Put together, here's the measurement stack that actually answers the AI question honestly:

• GA4 custom channel group for the AI traffic that clicks through, with the trend over time. (This guide.)
• Search Console for the Google side, including the organic movements that hint at AI Overviews activity on your queries. (Set up in the foundations guide.)
• An AI visibility audit for the impression side: whether the engines cite you, mention you, or leave you out of the answer entirely, click or no click. You can start on this half yourself by tracking your brand mentions in AI search.

The first is something you can and should set up yourself this week. It moves you from "no idea" to a real, trended number, and that alone changes the conversation. Just hold it with the right expectation: it's the floor of your AI footprint, the part visible enough to count. The influence above that floor is real too, and measuring it is a separate, deliberate exercise.

Get the countable part in place first. Then you can have the more interesting argument about the part that doesn't click. That's the work, and that's Citable.

Frequently asked questions