LLM traffic: how to track ChatGPT and AI referrals in GA4

To track LLM traffic in GA4, start with the default AI Assistant channel, which groups visits from sources such as ChatGPT, Gemini, Deepseek, Copilot and Grok. It excludes Google's AI Overviews and AI Mode, which stay in Organic Search. For your own definition, build a custom channel group with a regex for the assistants you care about, then analyse it in an Exploration.

What counts as LLM traffic in GA4?

LLM traffic is a visit that arrives on your site after someone clicks a link inside an AI assistant's answer. In GA4 it is ordinary referral traffic with a recognisable source, and Google now gives part of it a channel of its own.

The distinction matters because an AI assistant can mention or summarise your page without anyone clicking. GA4 only sees the click. So AI referral traffic is a floor, not a measure of how often you are cited. We treat it as one signal alongside others, which is the same caution we apply in our piece on AI share of voice.

It is also worth being clear about what GA4 cannot do here. It records a session and where that session came from. It does not tell you which question the person asked the assistant, or what the answer said about you. Anything you read about prompts has to come from somewhere else.

Where does ChatGPT referral traffic show up in GA4 by default?

In the AI Assistant channel of the default channel group. Google's default channel group page defines AI Assistant as the channel by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok.

The same page lists two ways a session lands there. One is a medium that exactly matches “ai-assistant”, which is what you would set yourself in a tagged link. The other is automatic: the medium is set to “ai-assistant” and the campaign to “(ai-assistant)” if the referrer matches a list of AI Assistants that Google maintains. That second rule is why ChatGPT referral traffic can appear in this channel without any setup on your side.

Two limits come with that convenience. Google says default channel groups can't be edited, and that the definitions reflect Analytics' current view of each channel and may evolve as the market changes. You do not control which assistants are on the list, and the list can change under you. If a referrer is not on it and matches no other rule, the page says the session can end up as Unassigned.

To see the channel, open the traffic acquisition report and look at the session default channel group. If AI Assistant is not there, GA4 has not seen a qualifying session in the date range you picked. That is a finding in itself, and we come back to it below.

Why Google's AI Overviews are not in the AI Assistant channel

Because Google counts them as search. The same definitions page says the AI Assistant channel excludes Google's AI Overviews and AI Mode, and that Organic Search includes them.

In practice this means a click from an AI Overview and a click from a classic blue link sit in the same channel, and the default report gives you no way to tell them apart. If your question is “how much of my AI search traffic comes from Google's AI features?”, the default channel group will not answer it. Do not read a quiet AI Assistant channel as proof that AI search sends you nothing; part of it may be inside Organic Search.

For Google's AI surfaces, we pair GA4 with a separate check of whether the site appears in them at all, which we cover in how to check if your site appears in AI Overviews.

How do you build a custom channel group for AI referral traffic?

Create a custom channel group with its own AI channel, define it with a regex, and place it above Referrals. Google's custom channel groups page sets out this exact use case, and adds that custom channel groups can be applied to your reports retroactively, so you see past data too.

The steps we follow:

1. Create a new custom channel group for AI referral traffic.
2. Add a new channel and give it a clear name. Google suggests something like “AI assistants”.
3. Set its condition to a regex matching the AI assistant URLs you want to include. Google's page gives ChatGPT, Gemini, Microsoft Copilot, Claude and Perplexity as examples of popular assistants.
4. Reorder the list so the AI channel sits above Referrals. Google notes you can also place it above other channels, such as Organic Search, if needed.
5. Save, then switch a traffic acquisition report to the new group and compare it with the default AI Assistant channel for the same dates.
6. Diary a review. Google advises updating the regex if the URLs or the list of assistants you want to measure change.

The ordering step is the one people miss. Google states that traffic is included in the first channel whose definition it matches, given the current order of channels in the group. Leave your AI channel below Referrals and every AI visit will match Referrals first, and your new channel will read zero.

On the regex itself, we do not guess hostnames. We open the session source dimension for the last few months, filter for the assistant names above, and copy the exact source values the property has actually recorded into the pattern. That way the regex describes your data rather than our assumptions about it.

How do you analyse LLM traffic in an Exploration?

Use a free-form Exploration with your AI channel as a dimension. Google's Explorations guide describes Explorations as giving you access to data and techniques that aren't available in reports, and says the default technique for a new blank exploration is the free form table.

1. Create a blank exploration and name it after the question, for example “LLM traffic by landing page”.
2. Import dimensions: your custom channel group, session source, and landing page.
3. Import metrics: sessions, engaged sessions and key events, or whatever you already report on.
4. Put landing page in rows and the AI channel as a filter, so you see which pages assistants actually send people to.
5. Add session source as a second row dimension to split ChatGPT from Perplexity and the rest.
6. Duplicate the tab and swap landing page for date to watch the trend.

Three constraints from Google's page are worth knowing before you build. You can apply up to 20 dimensions and 20 metrics, and up to 10 segments, per exploration. Explorations may use sampled data if more than 10 million events are part of a query. And by default, properties retain 2 months of data, which limits how far back an Exploration can look. If you want a longer history, Google says you can change it in Admin, under Data collection and modification, in Data Retention.

Session source or first user source: which one to read?

For LLM traffic, read session-scoped dimensions first. Google's page on traffic-source scopes explains the difference, and the choice changes what your numbers mean.

Google adds that user-scoped and session-scoped dimensions use the paid and organic channels last click model, and are unaffected by changes to the attribution model. That makes them stable for a monthly report. Event-scoped figures move if someone changes the attribution setting, so note the model whenever you quote them.

What to check when LLM traffic reads zero

Check that assistants can reach your pages before you question the tracking. An assistant cannot link to a page its crawler was refused, and GA4 will faithfully report the zero.

What we have seen: on one SEO Autopilot engagement, a US healthcare analytics company on WordPress, the firewall returned 403 to OpenAI's and Anthropic's search crawlers on every page. The trigger was a generic bad-bot rule matching the substring “searchbot”. Eleven other crawlers passed. The client's monitoring tool reported all sixteen AI crawlers as accessible, because it read robots rules instead of sending requests.

A short checklist we run before trusting an empty AI channel:

• Robots rules and firewall responses both allow the AI crawlers you care about; read the rules and send real requests.
• The custom AI channel sits above Referrals in its channel group.
• The regex matches the source values your property has actually recorded.
• The date range sits inside your data retention setting.
• Organic Search has been considered for Google's AI Overviews and AI Mode clicks.

Our free AI visibility checker reads robots.txt rules for 11 AI crawlers and requests your homepage once with each crawler's user agent. That is the checker's own request, not the real crawler's, and it covers the homepage only, so treat it as a first look rather than proof. Inside SEO Autopilot, crawler access and AI referral tracking are part of the same monthly report, alongside the SEO KPIs we track.

Frequently Asked Questions