How to Track ChatGPT, Perplexity and Gemini Traffic in GA4

Start with the Honest Version of the Number

GA4 can measure some visits from ChatGPT, Perplexity, Gemini and other answer platforms. It cannot measure every AI-assisted visit.

When a platform passes a document referrer, GA4 can populate the session source and medium. When that information is removed by an app, browser, redirect, privacy control or copied link, GA4 has no reliable source to assign and may report the session as direct.

Your AI-referral report is therefore a confirmed lower bound. It is valuable precisely because it is observable, but it should never be presented as the complete influence of AI on demand.

Use Session-Scoped Dimensions

For traffic acquisition, use Session source, Session medium or Session source / medium. Google defines these as the source and method associated with the session.

Do not casually mix them with First user source, which describes how a person was first acquired. Someone may first discover you through Google, return through ChatGPT and convert through email. Each scope answers a different question.

For a weekly AI report, the useful combination is:

• Session source / medium
• Landing page + query string
• Sessions and engaged sessions
• Engagement rate
• Key events
• Total revenue, where configured correctly

Find the Sources That Already Exist

Open Reports → Acquisition → Traffic acquisition and change the primary dimension to Session source / medium. Search for obvious terms such as “chatgpt”, “perplexity”, “gemini”, “copilot” and “claude”.

Export the actual hostnames before creating rules. Platforms change domains and product paths; your own data is the best starting inventory.

A practical rule might include observed sources such as:

• chatgpt.com
• perplexity.ai
• gemini.google.com
• claude.ai
• copilot.microsoft.com

Treat the list as maintained configuration, not permanent truth. Review unmatched referrals monthly and document every change.

Build an Exploration Before Changing Channels

Create a free-form exploration with a segment where Session source matches regex for your confirmed AI domains. Add landing page, device category and country as breakdowns.

This lets you validate the logic without changing organisation-wide reporting. Check that:

1. the segment captures the sources you intended
2. ordinary search engines are not included by a loose expression
3. internal tools and test traffic are excluded where appropriate
4. key events are configured and firing correctly
5. the date range contains enough sessions to interpret responsibly

Once the segment is stable, reproduce the logic in a custom channel group if stakeholders need a permanent “AI Referral” line.

Measure Landing-Page Intent, Not Just Volume

A hundred AI sessions to a generic article and ten AI sessions to a high-intent service comparison are not equivalent.

Group landing pages into intent levels:

Use content groups or a maintained URL mapping so the analysis survives individual page changes.

Create a Small Funnel

For a service business, a useful AI-referral funnel might be:

1. AI-referred session
2. engaged session
3. service or case-study view
4. contact-form start
5. qualified enquiry

The last step often lives in the CRM rather than GA4. Join records through appropriate campaign and lead-source fields where consent and system design allow it. Avoid sending personal or sensitive data into analytics.

Handle Direct Traffic Properly

Google defines direct traffic as sessions without clear referral information. Do not reclassify a slice of direct traffic as AI just because direct increased after ChatGPT became popular. That would replace an unknown with an unsupported claim.

Instead:

• report identifiable AI referrals separately
• label them as a minimum
• track changes in direct traffic as contextual evidence only
• use customer surveys or CRM “How did you hear about us?” fields as a second signal
• compare branded search and assisted-conversion patterns without claiming causation

Multiple imperfect signals can support a conclusion. None should be disguised as complete attribution.

Use UTMs Where You Control the Link

You cannot add campaign parameters to links an independent AI system chooses to cite. You can use UTMs when you control the placement: a custom GPT, a managed assistant, a partner integration or an AI-focused campaign.

Use a documented convention, for example:

utm_source=custom_gpt&utm_medium=ai_assistant&utm_campaign=service_advisor

Keep these sessions separate from organic AI referrals. One is a managed campaign; the other is earned or algorithmic discovery.

The Weekly Dashboard

Report only what helps a decision:

Compare at least four weeks where volumes are low. A change from two sessions to four is 100% growth and almost no evidence.

Validate Before You Present It

Click a controlled tagged link from any assistant experience you manage. Confirm the real-time session, source, landing page and key event. Then inspect the next day's processed reporting.

Finally, add a note to every dashboard: “AI referrals include sessions where a recognised source was passed to GA4. Some AI-assisted visits cannot be identified and may appear in direct or other channels.”

That sentence protects the report from becoming more certain than the data.

Official Sources

• Google Analytics: Traffic acquisition report
• Google Analytics: Campaigns and traffic sources
• Google Analytics: Understanding direct traffic
• OpenAI: Publishers and developers FAQ