How to track AI referral traffic in GA4

You've heard that shoppers are starting to find stores through ChatGPT and Perplexity instead of Google. Fair enough. But the obvious next question is the hard one: can you actually see any of that in your own analytics? The answer is yes, partly. Some of it shows up clearly in Google Analytics 4 if you know where to look. Some of it hides. This post walks you through exactly how to find the visible part, and how to stay honest about the part you can't see.

No new tools, no tracking scripts to install. Just GA4, which you almost certainly already have, and about fifteen minutes.

The signal you're looking for: AI hostnames

When an AI assistant sends someone to your site by clicking a link, the browser usually passes along a referrer, the same way any other site does. And the referrers from AI assistants have recognizable hostnames. The main ones you'll run into are:

• chatgpt.com — ChatGPT (you may also see older or related OpenAI domains)
• perplexity.ai — Perplexity
• gemini.google.com — Google's Gemini assistant
• copilot.microsoft.com — Microsoft Copilot

That's the whole trick, really. If you can get GA4 to show you sessions whose source or referrer contains one of those, you've isolated your AI referral traffic. The assistants do add and rename domains over time, so rather than memorizing every exact subdomain, it's easier to filter on the recognizable word: chatgpt, perplexity, gemini, copilot.

Don't expect a big number. For most stores this traffic is still a thin slice next to Google. But it tends to be unusually high-intent, because the person already got a recommendation before they clicked.

Option A: a quick segment in an Exploration

The most reliable way to see this in GA4 is to build a Free Form Exploration. It gives you full control over what counts, instead of waiting on GA4's default channel groupings, which don't yet reliably bucket every assistant the same way.

Step by step

• In GA4, open Explore in the left sidebar, then start a Blank (Free Form) exploration.
• Set your date range up top. Pick a wide window, like the last 90 days, because the volume is small and a short window can look like zero.
• Under Dimensions, click the plus and add Session source / medium (and, if you want the raw referring URL, add Page referrer as well). Under Metrics, add Sessions, Engaged sessions, and if you have ecommerce set up, Conversions or revenue.
• Drag Session source / medium into the Rows field and your metrics into Values.
• Now create a filter. In the filter area at the bottom of the Tab Settings, choose Session source / medium, set the match type to contains, and enter chatgpt. Apply it.

You'll now see just the ChatGPT sessions. Repeat for the others, or use a filter condition that matches a regex like chatgpt|perplexity|gemini|copilot if you'd rather see them all in one table. Add a second dimension, like Landing page, and you'll see exactly which products or articles the AI is sending people to. That last detail is often the most useful thing in the whole report.

Option B: a reusable segment

If you'd rather flip this view on and off across reports, build a segment instead. In your Exploration, under Segments, click the plus and choose a Session segment. Add a condition on Session source (or Page referrer) with the match type contains, and add each AI hostname as an "OR" condition: chatgpt, then perplexity, then gemini, then copilot. Save it with a clear name like "AI assistants." Now you can drop that segment onto any Exploration and instantly compare AI-driven visitors against everyone else, on engagement rate, pages per session, or conversion.

Comparing segments is where this gets interesting. If your AI visitors convert at a different rate than your search visitors, that's worth knowing, and it's the kind of thing a raw referral list won't tell you.

The caveat nobody mentions: missing referrers

Here's the honest part. Not all AI traffic carries a referrer. When someone follows a link from inside a mobile app, or from a context where the referrer is stripped, the visit can land in GA4 with no source at all, which usually gets bucketed as Direct. That means the AI numbers you can see are a floor, not a ceiling. The real total is higher, and you have no clean way to recover the missing slice from GA4 alone.

So treat your GA4 AI numbers as directional. They prove AI traffic exists and they show you the trend over time, but don't present them as the complete count. If you see your "AI assistants" segment growing month over month, that's a real signal, even though the absolute number undersells reality.

This is also why analytics alone can't answer the bigger question of whether AI engines actually recommend you in the first place. GA4 only sees the people who already clicked through. It can't tell you how often ChatGPT named a competitor instead of you, with no click at all. For that side of the picture you need to test the assistants directly. If you'd rather not do that by hand across four tools, you can run a free AI visibility audit and we'll check whether the major engines mention your store and which competitors they name instead.

What to actually do with the numbers

Once you can see AI referrals, the useful moves are simple. Watch the trend, not the daily noise. Note which landing pages the assistants favor, because those are the pages that are already AI-friendly, and you can learn from them. And pair this with the supply side: if the traffic is thin, the fix usually isn't in GA4, it's in whether AI can read and trust your site at all.

If your numbers are stubbornly flat, two common culprits are worth ruling out. One is that AI crawlers can't reach your pages, which is more common than people expect; our guide on whether your store is blocking AI crawlers covers how to check that in a couple of minutes. The other is that you're confusing falling Google clicks with falling demand, when AI Overviews are quietly answering shoppers before they reach you; that's the subject of why your Google traffic can drop while rankings hold steady.

Measurement is the easy half. It tells you where you stand today. The harder, slower half is earning the mentions, reviews, and clean product data that make AI name you in the first place. But you can't manage what you can't see, so start by getting GA4 to show you the AI traffic you already have. It's a fifteen-minute setup, and once it's in place you'll never have to guess about this again.