How to see AI referral traffic in GA4

Most guides on this topic open by telling you to build a custom channel group with a long regex of AI domains. Before you do that, check your Traffic acquisition report, because Google now ships the channel.

Google's default channel group documentation lists an AI Assistants channel, described as the channel by which users arrive from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok. Checked 2026-08-06. Under the hood, Google sets the medium to ai-assistant and the campaign to (ai-assistant) when the arriving referrer matches its list, and the channel is defined as medium exactly matching ai-assistant.

That is a genuine improvement and it comes with a catch worth understanding before you report a number to anyone.

The catch, stated first

The same documentation says the AI Assistants channel excludes Google's AI Overviews and AI Mode, and that Organic Search includes them.

So a visitor who clicked a link inside an AI Overview is in your Organic Search row, mixed in with ordinary blue-link clicks, and there is no default channel that separates them. Your "AI traffic" number in GA4 is assistant traffic, not AI-influenced traffic. Those are different things and the gap between them is not small.

Say that out loud whenever you present the figure. It is the difference between a measurement and a talking point.

Do the task

1. Traffic acquisition, one dimension change

Go to Reports, then Acquisition, then Traffic acquisition. The report opens with Session default channel group as the primary dimension. Look for the AI Assistants row.

If it is not there, you either have no matching sessions in the date range or your property is showing a primary channel group you have customized. Google's documentation notes that any custom channel group can be promoted to be your property's editable primary channel group, in which case your rules are what the report is using, not Google's.

To see which assistant, click the dropdown at the top of the first column and switch to Session source or Session source / medium. Then use the search box above the table.

One detail that trips people up. Because GA4 rewrites the medium to ai-assistant for referrers it recognizes, a row may read chatgpt.com / ai-assistant rather than chatgpt.com / referral. If you built old reports that filter on referral as the medium, they will now miss this traffic. Check any saved segment or filter that assumes it.

2. Build a custom channel group only if you need your own list

There is a real reason to do this: Google's list of recognized assistants is Google's, and it is not published in full. If you need a list you control and can audit, build one.

Per Google's custom channel groups documentation, checked 2026-08-06:

Go to Admin, then under Data display, click Channel groups. You need Editor access or above on the property. Click Create new channel group, which starts you with a copy of the default group. Add a channel, name it, and give it a condition on Source using matches regex. Google's own worked example for this is an AI assistants channel, and it names ChatGPT, Gemini, Microsoft Copilot, Claude, and Perplexity as the assistants to include.

Then the step everyone forgets, which Google calls out explicitly: reorder so your channel sits above Referral. Traffic is assigned to the first channel whose definition it matches, so a channel underneath Referral catches almost nothing.

Three limits worth knowing before you commit. Standard properties get 2 custom groups in addition to the predefined group, 360 properties get 5, and each group holds up to 50 channels. Custom channel groups apply retroactively, which is why this is worth doing even for a period that has already passed. And they are not available in the BigQuery export schema, so a warehouse pipeline needs its own logic.

3. Explorations, for the part reports will not give you

Reports tell you how many. Explorations tell you which pages, which is the part you can act on.

Per Google's Explorations documentation, checked 2026-08-06:

Click Explore in the left navigation, then create a blank exploration. The default technique for a blank exploration is the free-form table.

In the Variables panel on the left, click the plus next to DIMENSIONS and import Session source, Session medium, and Landing page + query string. Then the plus next to METRICS and import Sessions, Engaged sessions, and Key events.

In Tab Settings, drag Landing page into ROWS and Session source into COLUMNS, and drag Sessions and Key events into VALUES. Then add a Filter on Session source using matches regex with the assistants you care about.

You now have the useful view: which specific pages assistants are sending people to, and whether those people do anything.

Limits to know, per that same documentation: 20 dimensions and 20 metrics per exploration, 10 filters per tab, and 10 segments per exploration. An exploration may be based on sampled data if more than 10 million events fall inside a particular query. The indicator in the corner turns from green to yellow when an exploration is subject to either sampling or a data threshold, so a yellow icon on its own is not proof that you were sampled; open the tooltip. Export formats are Google Sheets, TSV, CSV, PDF, and PDF (all tabs), and exporting to Sheets, TSV or CSV sends all the data in the visualization rather than only the rows on screen.

4. Write down what the number cannot see

This is a step, not a footnote. Do it while you are in the report, not after someone has quoted the figure in a meeting.

Answers that never produce a click. An assistant that answers the question completely sends nothing at all. This is the same mechanic as zero-click search and it is the largest invisible category by far.

Clients that send no referrer. Sessions from native mobile apps and desktop clients frequently arrive with no referrer at all, in which case GA4 has nothing to match and the session lands in Direct. Same for anyone who copies a URL out of an answer and pastes it into a new tab.

AI Overviews and AI Mode. Covered above. They sit in Organic Search, per Google's own definition.

Small numbers. Search Engine Land's audit of 15 domains covered nearly 2 million organic monthly sessions and 7,500 ChatGPT referral sessions across that whole set, which works out at roughly one ChatGPT referral per 265 organic sessions. That ratio describes those 15 domains on that date, not your site, and it counts one assistant rather than all of them. Take it only as an order of magnitude: where the numbers sit in that region, a dramatic month-over-month percentage swing in assistant traffic can be a handful of sessions. Report absolute numbers alongside percentages.

Google's list changes. Google's documentation says its channel definitions may evolve as the market changes, and it tells you to update your own regex if the assistants you track change. Neither list is stable.

What this measures, and what it does not

GA4 answers one question well: how many people arrived, from which assistant, onto which page, and what they did next.

It cannot answer whether an engine could read the page it did not send anyone to. It cannot tell you which of your pages an assistant skipped, or which competitor it named instead. Referral counts are downstream of readiness; they are a consequence, and a partial one.

The complementary measurements are your citation rate across engines, which asks what engines actually say when prompted, and AI crawlers in your server logs, which asks whether the fetch even happened.

The damaging admission

GA4 is free, it is probably already installed, and everything in this guide takes about twenty minutes. Citedon does not connect to your GA4 and does not read it. If a traffic number is all you want, you are done here and you do not need us.

We will also say the harder thing: assistant referral traffic is a weak proxy for readiness and a worse target to optimize for. A number that moves for reasons you cannot attribute is not a control, it is a mood. Treat it as context around measurements you can act on.

Check what the engines can actually read

Pull the number, write the caveats next to it, then go look at whether your pages are legible to the engines in the first place.

Run a free scan to see how ChatGPT, Perplexity, Gemini, and Claude read your key pages today. The scan diagnoses any site. Automated apply through the connected Citedon plugin is WordPress only, and nothing in this GA4 setup requires a plugin at all.