How to track AI traffic in GA4: setup, validation and limits

The 60-second answer

To track AI traffic in GA4, start with the new AI Assistant channel. When GA4 recognises an AI-assistant referrer, it assigns the medium ai-assistant and places the visit in that channel. Google announced this change on 13 May 2026 (Google Analytics release notes).

That is useful, but the channel is not a complete measure of AI influence. Some identifiable visits may still carry a different medium. Visits without usable source information can appear as Direct. Google’s AI Overviews and AI Mode remain within Organic Search in GA4.

Use this measurement process:

1. inspect Session source / medium, not only the channel total;
2. create a custom channel based on verified source values;
3. validate the rules against Referral, Unassigned and Organic Search;
4. measure engagement, key events and qualified leads;
5. report the result as identifiable AI referral traffic, not total AI traffic.

A channel group improves classification. It cannot recover data that was never collected or measure a recommendation that produced no website visit.

1. Decide what “AI traffic” means before opening GA4

“AI traffic” can describe several different things:

These measurements are related, not interchangeable. Generative engine optimisation, or GEO, concerns whether AI systems can find, understand and cite a company. Referral traffic covers only users who click.

2. Audit your existing ChatGPT and AI referral traffic

Before building a GA4 custom channel group, inspect what the property already records.

Open:

Reports → Acquisition → Traffic acquisition

Use Session source / medium as the primary or secondary dimension. The Traffic acquisition report uses session-scoped dimensions, which describe how a particular session began. User acquisition uses first-user dimensions and answers a different question: how the person was initially acquired (Google Analytics: traffic-source scopes).

Search the source rows for assistant names and domains already present in your data. Review at least:

• the default AI Assistant channel;
• Referral;
• Unassigned;
• Organic Search;
• Direct, as a limitation rather than a source you can safely reclassify.

Google’s current default definition places a visit in AI Assistant when its medium is exactly ai-assistant. The current description names sources such as ChatGPT, Gemini, DeepSeek, Copilot and Grok, while explicitly excluding Google AI Overviews and AI Mode (Google’s default channel definitions).

Do not assume every row from one assistant has the same source-and-medium combination. Source and medium are separate fields, so inspect the combinations in your property rather than relying on the channel name.

Create a small audit table:

The rows in your account are the source of truth for the rule you build.

3. Build a GA4 custom channel group for identifiable AI referrals

Google provides an AI-assistant example in its custom-channel documentation. Rules can use Source, Medium and Source platform, and the first matching channel wins (Google Analytics: custom channel groups).

To create the group:

1. Open Admin → Data display → Channel groups.
2. Select Create new channel group.
3. Add a channel named Identifiable AI referrals.
4. Set the condition to Source matches regex.
5. Enter a pattern based on source values you have verified.
6. Move the new channel above Referral.
7. Save the group and test it as a secondary dimension.

A cautious starter pattern could look like this:

^(www\.)?(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|deepseek\.com|grok\.com|you\.com)$

This is a template, not a universal list. Use only verified source values. Avoid broad fragments such as ai, gpt or google, which can match unrelated traffic.

Placing the custom channel above Referral allows it to claim matching AI referrals first. Do not move it above Organic Search until you have tested that the rule cannot absorb ordinary search traffic.

4. Understand the difference between retroactive reports and the primary channel group

This GA4 detail is easy to miss.

Google states that custom channel groups can be applied retroactively to reports. However, if you make one the property’s primary channel group, that primary definition populates reporting from that point forward.

The safer sequence is:

1. keep the Google default group unchanged as a reference;
2. use the custom group as a secondary dimension;
3. compare both definitions across the same date range;
4. document the rule and approval date;
5. change the primary group only if the organisation needs one shared custom definition.

Custom channel groups are not available in the GA4 BigQuery export schema. Teams using BigQuery must reproduce the rules in their queries.

5. Validate the custom AI channel before reporting it

Run five checks before reporting:

Coverage

Confirm that every verified AI source row reaches the custom channel.

False positives

Check that ordinary search, referral and campaign traffic has not been pulled in.

Priority

Confirm that the AI channel sits above Referral. GA4 assigns traffic to the first matching channel in the ordered group.

Missing information

Review Unassigned and (not set) rows. Unassigned appears when no channel rule matches, while Direct represents traffic without a clear referral source (Google Analytics: Direct traffic).

If the source itself is absent, a regex cannot recover it. Do not estimate an “AI share of Direct” and present it as observed traffic.

Change control

Save the domain list, rule, test result, owner and review date. A channel without an owner will become stale.

6. Report business value, not only AI session volume

Once classification is stable, report buyer quality:

• identifiable AI referral sessions;
• engaged sessions and engagement rate;
• key events, such as a qualified form or booked call;
• session key-event rate;
• landing pages receiving AI referrals;
• source-level performance;
• qualified opportunities and revenue where CRM matching is possible.

Compare landing pages and business outcomes, not only channel volume. A few service-page visits may matter more than many low-intent article visits.

Use consistent labels in reports:

• Default AI Assistant for Google’s maintained definition;
• Identifiable AI referrals for your verified source-based group;
• AI visibility for mentions and citations measured outside GA4;
• AI-influenced pipeline only when CRM or customer evidence supports the claim.

This naming prevents a classification improvement from being presented as new demand.

7. What even a well-built GA4 AI channel cannot measure

These are AI traffic attribution limits, not problems a larger regex can solve.

AI visits with no identifiable source

Google classifies (direct) / (none) when traffic has no clear referral source. Some of those visitors may have used an assistant, but GA4 does not provide evidence to identify which ones.

Google AI Overviews and AI Mode as a separate GA4 channel

Google includes these experiences in Organic Search rather than AI Assistant. Do not create a source rule that captures all google / organic; it would combine ordinary search and generative-search visits.

Search Console now has a separate generative-AI report for impressions in AI Overviews and AI Mode. It is rolling out to a subset of properties and adds visibility evidence, not GA4 referral attribution (Google Search Console documentation).

Recommendations without clicks

An assistant may mention the brand, answer the question and end the journey without sending a visit. Website analytics cannot measure that exposure.

Influence before a later return

A buyer may discover the company in an assistant, return through branded search and convert later. Use CRM source questions, sales notes and attribution analysis to investigate.

8. Use a four-layer AI measurement model

Review the layers together, but do not merge them into one unsupported score. The system should separate observed visits, observed visibility and inferred commercial influence.

Conclusion: improve classification without overstating attribution

The GA4 AI Assistant channel is a useful default. A verified custom channel can provide a more consistent view of AI referrals across historical reports.

Neither one measures total AI influence.

Start with Session source / medium, build the rule from observed sources, test for false positives and report the result as identifiable AI referral traffic. Then combine it with Search Console, citation monitoring and CRM evidence.

The strongest report is not the one with the largest AI number. It is the one whose definition, limits and business consequences can be explained.

Frequently asked questions