AI search attribution in GA4: how to track ChatGPT, Perplexity and Copilot referrals

GA4 now tracks most AI assistant traffic by itself. In mid-2026 Google added an AI Assistant channel to the default channel group, and it covers visits from ChatGPT, Gemini, Copilot, DeepSeek and Grok. Perplexity and Claude are not on Google's list at the time of writing, so their visits still land in Referral. Clicks from Google's AI Overviews and AI Mode are counted as Organic Search. A custom channel group fills the gaps.

This guide covers what the built-in channel does, what it leaves out, how to set up a channel that catches everything, and how to attribute conversions to AI search.

#What the built-in AI Assistant channel does

Google's default channel group documentation describes the rule. When the referrer of a visit matches Google's list of AI assistants, GA4 sets the session medium to ai-assistant and the campaign to (ai-assistant). Any session with that medium goes into the AI Assistant channel.

To see it:

1. Open Reports, then Acquisition, then Traffic acquisition.
2. Find the AI Assistant row in the default channel group.
3. Add Session source as a second dimension to see which assistant each visit came from.

Two limits to know about:

• It starts from when it was introduced. Sessions recorded earlier stay where they were, usually in Referral.
• It depends on the referrer. A visit that arrives with no referrer cannot be matched.

#Where each source lands

So a default report undercounts. Perplexity alone can be a meaningful share of AI visits for a B2B site, and it is sitting in a different row.

#A custom channel group that catches the rest

A custom channel group lets you define your own channel. It has one more advantage: it applies to your historical data, so you can see the months before the built-in channel existed.

#Step 1. Create the group

In GA4, go to Admin, then Data display, then Channel groups, and choose Create new channel group. Menu names change from time to time, so look for "Channel groups" if yours differs. Starting from a copy of the default group keeps all the standard channels.

#Step 2. Add an "AI search" channel

Give the new channel two conditions joined with OR:

Medium exactly matches: ai-assistant

Source matches regex:
chatgpt\.com|chat\.openai\.com|perplexity|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|deepseek|grok|chat\.mistral\.ai|you\.com|phind\.com

The first condition picks up everything Google already classifies. The second adds the assistants Google leaves out, and it also catches older sessions from before the built-in channel.

#Step 3. Move it above Referral

GA4 checks channels from top to bottom and uses the first match. Drag the new channel above Referral, or Perplexity visits will keep falling into Referral.

#Step 4. Use it in reports

In the traffic acquisition report, open the dimension menu and pick your custom channel group. New groups can take a day or so to fill in.

#Step 5. Check it

Ask an assistant a question that cites your site, click the link, then look at the Realtime report. The visit should appear under your new channel.

#Three things that confuse the numbers

#ChatGPT tags its own links

ChatGPT adds utm_source=chatgpt.com to many of the links it shows. That is useful: some visits name their source even when no referrer is sent. In older data these often appear as chatgpt.com / (not set) and sit in Unassigned. The source condition in the custom channel above picks them up.

Do not add that tag to your own links. It would label every visitor who uses the link as ChatGPT traffic.

#Google's AI answers are not a separate source

A click from an AI Overview or from AI Mode arrives from Google, like any other search click. GA4 records it as google / organic. Google's documentation says the AI Assistant channel excludes them.

Search Console is the better place to look. Google says that appearances in AI Overviews and AI Mode are included in the Performance report totals under the Web search type. They are not shown separately, but there is a clue in the query list: long, conversational queries written like a prompt tend to come from AI-style search.

#Small numbers swing a lot

For most sites, AI visits are still a small share of the total. With a few dozen sessions a month, one sale moves the conversion rate enormously. Look at the trend over months, and be slow to draw conclusions from a single week.

#Attributing conversions to AI search

There are three views, and each answers a different question.

Last-touch. Key events and revenue for sessions in your AI channel. This is the easy one, and it undercounts. People often learn about a company from an assistant and come back later by typing the address or searching the name. Those visits are credited to Direct or Organic Search.

First-touch. In the User acquisition report, use the first user source. This shows people whose first visit came from an assistant, whatever they did afterwards. It is closer to the question "did AI search introduce them?"

Ask them. Add a "How did you hear about us?" field to your contact and sign-up forms. It is the only method that captures people who saw you in an answer and never clicked. The answers are imperfect and still valuable.

Several industry reports suggest that visits from AI assistants convert at a higher rate than average. The figures vary widely and the samples are small, so compare the conversion rate of your AI channel with Organic Search in your own data before you rely on it.

#What to put on a dashboard

Build these once in GA4's Explore section and save them.

• AI sessions per week, as a line over the last three to six months. This is the trend.
• AI sessions by source. Which assistants send visitors.
• AI sessions by landing page. Which of your pages are being cited and clicked.
• Key events from the AI channel, next to the same figure for Organic Search.

The landing page view is the most useful for deciding what to write next. Pages that already attract AI visits are worth deepening.

#Visits are only half the picture

Analytics measures clicks. It cannot tell you whether an assistant mentioned you in an answer that nobody clicked.

For that, run the questions your buyers ask and record whether you are cited, mentioned or absent. Then read the two measures together:

• Mentions rising, visits flat. You are being named but people are not clicking. Check that the cited pages have clear titles and that the answer leaves a reason to visit, such as a tool, a price list or a template.
• Visits rising, mentions flat. The assistants themselves are sending more traffic. Good news, and not of your making.
• Both flat. Go back to the basics: crawler access, indexing and pages that answer a question directly. Our guide to getting cited by ChatGPT has a checklist.

For ways to track mentions, see how to track LLM citations.

#If you use another analytics tool

The same idea applies to Plausible, Fathom, Matomo or server logs: group the known assistant hostnames into one channel or segment. Only the place where you define the rule changes.

#If you would like help

The setup above takes under an hour and you can do it yourself. If you would like a wider review of how AI assistants find and describe your site, our AI search visibility audit is a one-week review for a fixed $1,800. It covers crawler access, indexing, structured data and content structure, and it records a baseline of where assistants mention you today.

To talk it through first, book a free 20-minute call.

#Bottom line

Use the built-in AI Assistant channel for a quick view, and add a custom channel group so that Perplexity, Claude and your older data are counted too. Remember what analytics cannot see: Google's AI answers are inside Organic Search, app visits arrive as Direct, and a mention with no click leaves no trace. Track visits in GA4, track mentions by asking the questions yourself, and read the two together.