Track LLM traffic in Google Analytics 4: 5 methods compared

To track LLM traffic in Google Analytics 4, start with the AI Assistant channel that Google began rolling out in mid-May 2026. Then add a custom channel group that matches ChatGPT, Perplexity, Claude, Gemini and Copilot by source. The channel group also counts sessions from before the update.

This guide is for marketers, site owners and agencies who report from GA4 and need AI referrals on a line of their own. It covers where the native channel falls short, a regex you can paste into any report, step-by-step setup for four manual methods, and the LLM traffic GA4 still cannot see.

Why LLM traffic deserves its own line in GA4

A PROS guide on the subject reports that traffic from large language models such as ChatGPT and Gemini grew rapidly through 2025. For most of that time, GA4 had no bucket for it. These visits landed in Referral, or in Direct when no referrer came through, mixed in with forum links and partner sites.

That mix hides a valuable audience. A Volt analytics guide cites Semrush data from July 2025 showing that LLM visitors convert at 4.4 times the rate of organic search visitors. In the same data, ChatGPT visitors viewed an average of 2.3 pages per session, against 1.2 for organic searchers.

These visitors arrive primed. They have just read an answer that cited your page, so they already trust the topic and often the brand. Give AI referrals their own line and you can see which pages the assistants link to, how those visitors engage, and whether they reach your key events.

How GA4 labels AI traffic today

The native AI Assistant channel

GA4 puts every session into a channel, which is a named bucket built from rules on Source, Medium and Campaign. It checks the rules from top to bottom and assigns the session to the first one that matches. Anything left unmatched falls into Referral, Direct or Unassigned.

From mid-May 2026, Google began rolling out a native channel for AI. When a session comes from a recognized assistant, GA4 writes "ai-assistant" into the session's Medium. The Default Channel Group then files it under a new channel called AI Assistant, next to Organic Search and Referral in your standard reports. Once your property has the update, there is nothing to set up.

What the native channel does not do

GA4 applies the label as sessions come in. It does not go back and relabel sessions that were filed under Referral before your property got the update. A full-year trend or a before-and-after comparison still means building it yourself.

Part of the coverage is also unknown. Google has named ChatGPT, Gemini and Claude as recognized sources but has not published the full list. For now, Perplexity, Microsoft Copilot, Grok, DeepSeek and Mistral are in a gray zone.

5 ways to track LLM traffic in Google Analytics 4

Each method answers a different question. The native channel covers day-to-day tracking, and the other four fill the gaps in history and coverage.

Method	Best for	Includes data from before May 2026?	Setup effort	Main limitation
Native AI Assistant channel	Ongoing tracking of recognized assistants with no setup	No, labels start when your property got the update	None	Google has not published its list of recognized sources
Regex filter in Traffic acquisition	A quick spot check of AI sessions, old or new	Yes	Low	The filter disappears when you close the report
Exploration with a filter or segment	Comparing assistants and the landing pages they cite	Yes	Medium	Lives in Explore, not in the standard reports
Custom channel group	One AI channel in every report and exploration, history included	Yes	Medium	The regex needs occasional upkeep
Custom report in the Library	A permanent AI traffic report inside the Reports menu	Yes	Medium	Needs at least Editor access to the property

For most properties, the strongest setup pairs the native channel with a custom channel group. The group catches what Google's list misses and keeps your history in the same view.

The regex behind every manual method

All four manual setups use one regular expression matched against the source field. This version covers the major assistants named in current GA4 guides:

.*(chatgpt|openai|perplexity|claude\.ai|anthropic|gemini\.google|copilot|deepseek|meta\.ai|grok|mistral|you\.com|poe\.com|phind).*

Three choices in it are deliberate:

• The .* on each side lets the pattern match a whole source such as chat.openai.com or chatgpt.com, and not only the brand word inside it.
• There is no catch-all for .ai domains. Some published patterns include .*\.ai$, which counts every site on a .ai domain, AI tool directories and SaaS blogs included. Naming the assistants one by one keeps those false positives out.
• There is no bare "gpt". On its own it also matches directories and review sites that have GPT in their domain.

Match it against Session source for session-level reports, or First user source to see which users an assistant brought in first. Keep the list current. The University of Illinois brand team's GA4 training suggests updating its pattern monthly as new tools appear.

Step-by-step setup for the four manual methods

Method 1: A quick filter in Traffic acquisition

1. Go to Reports > Acquisition > Traffic acquisition.
2. Click Add filter at the top of the report.
3. Pick Session source as the dimension and "matches regex" as the match type.
4. Paste the regex and click Apply.

The table now shows sessions, engagement and key events for AI referrals alone. The filter disappears once you leave the report, so it suits spot checks better than ongoing monitoring.

Method 2: An Exploration you can come back to

1. Open Explore and start a Blank exploration.
2. Add the dimensions Session source / medium, Landing page + query string and Page referrer.
3. Add the metrics Sessions, Engaged sessions, Engagement rate and Key events.
4. Add a filter on Session source using "matches regex", or build a session segment with the same condition.

A filter narrows dimensions and metrics. A segment narrows the users, sessions or events themselves, and segments exist only in Explorations. Break the results down by landing page and you see exactly which content the assistants link to. Split them by source and you see whether ChatGPT visitors behave differently from Claude visitors.

Method 3: A custom channel group for LLM traffic in GA4

1. Open Admin (the gear icon), then Channel groups under Data display.
2. Create a new channel group, or copy the default group so you keep your existing channels.
3. Add a new channel and give it a clear name, such as AI Traffic.
4. Set the first condition to Source matches regex, and paste the pattern.
5. Add an OR condition: Medium exactly matches ai-assistant.
6. Save the channel, click Reorder and drag AI Traffic above Referral.
7. Save the group and make it your primary channel group.

Volt recommends the OR condition as a way to cut maintenance. Your regex catches platforms Google has not confirmed, and the medium rule picks up anything Google adds to its list later. The order matters because GA4 assigns each session to the first channel it matches, so a channel left below Referral stays empty.

The group shows up as a dimension in standard reports, Explorations and segments. It also covers your historical data, going back to when the property started collecting.

Method 4: A permanent report in the Library

1. Open Reports and click Library at the bottom left. You only see it with Editor access or higher.
2. Create a new detail report from the Traffic acquisition template.
3. Keep the dimensions you need, including your new channel group if you built one.
4. Add a report filter on Session source, "matches regex", with the pattern.
5. Save the report and name it.
6. Edit the Life cycle collection and drag the report under Acquisition.

What GA4 still misses

Even with every method in place, some of your AI traffic arrives with no clue about where it came from.

Visit type	How it usually appears in GA4	Attribution reliability	What you can do
Link clicked in Claude or Gemini on the web	Correct source, caught by the regex or the native channel	High	Nothing extra needed
Link clicked in ChatGPT or Perplexity	Often a "(not set)" medium	Mixed	Keep your regex on Source, where the platform name still shows
Link opened in an AI mobile app's in-app browser	Direct, because the referrer is often stripped	Low	Watch Direct traffic to the pages AI answers cite
URL copied from an answer and pasted into a browser	Direct	None	Treat it as unmeasurable in GA4
Click from a Google AI Overview	Google organic search	Cannot be separated in GA4	Check Search Console instead

AI Overviews sit inside Google Search, so their clicks arrive as ordinary organic traffic. Google's guidance for site owners on AI features, published on Google Search Central, counts those clicks in the standard web search reporting in Search Console.

So your GA4 AI figure is the verified minimum. You can cut the noise around it by tagging every link you control, such as newsletter, social and partner links, with a free UTM builder. The less unexplained Direct traffic you carry, the easier it is to spot an AI-driven lift on a cited page.

Turning AI referral data into decisions

Once AI Traffic has its own channel, review it on a schedule. A monthly check suits most sites.

Start with landing pages. The pages that draw AI sessions are the ones assistants trust enough to cite. Compare their engagement rate and key events with the same pages' organic sessions. If AI visitors convert better, as the Semrush figures suggest they often do, those pages are worth updating and linking to from more of your site.

Next, look at the numbers by source. They show which assistants cite you at all, which tells you where your content is surfacing and where it is not. Put the AI line in the same monthly scorecard you already use to prove your SEO is working. That way stakeholders see search and AI referrals side by side in one view.

FAQs

Does GA4 track ChatGPT traffic automatically now?

Partly. Since Google began rolling out the AI Assistant channel in mid-May 2026, recognized sessions get the medium "ai-assistant", and Google names ChatGPT as one of those sources. Sessions from before your property received the update stay in Referral.

Why does some ChatGPT traffic show as Direct or "(not set)"?

ChatGPT and Perplexity pass referrer data less consistently than Claude and Gemini. Clicks inside mobile app browsers often lose the referrer. A URL copied from an answer and pasted into a browser always arrives as Direct.

Will a custom channel group change my historical data?

It changes how reports group sessions, past ones included, without altering the raw data. That makes it the best way to see AI traffic from before the native channel existed.

Should I filter on Session source or First user source?

Use Session source to count every visit an assistant sends. Use First user source to see how many new users an assistant brought to your site first.

How often should I update the LLM regex?

Monthly is a safe default, because new assistants keep appearing. Adding the "Medium exactly matches ai-assistant" condition to your channel group cuts the upkeep, since Google's own detections then flow in automatically.

Build your AI Traffic channel group today

Open GA4 Admin and create the AI Traffic channel group using the regex and the ai-assistant condition above. Once the sessions start flowing in, see how to measure what pays to turn that channel into a number worth reporting.

Resources / References

1. Google Search Central

Sources

• How to Track LLM Referral Traffic in GA4 for Airlines - PROS
• What % of your site traffic is from LLM's in GA4?
• How to Track LLM (AI) Traffic in Google Analytics
• How to Track LLM Referral Traffic in GA4
• Setting Up Google Analytics to Identify AI-Generated Traffic