How to Track AI Traffic in GA4: From ChatGPT Referrals to SaaS Signups

The short answer

Start in GA4's Traffic acquisition report.

Review the Session default channel group dimension for AI Assistant, then inspect Session source / medium to understand the recorded sources.

Google now documents AI Assistant as a default channel. It includes traffic from recognized assistant sources.

Google's AI Overviews and AI Mode are excluded from that channel. They are included in Organic Search.

This is why filtering only for the Referral channel is no longer a complete starting point.

Also remember the measurement boundary:

GA4 can help report identifiable visits and events on your website. It does not show every brand mention, every generated answer, or every buyer influenced by an assistant before visiting later.

What this report can—and cannot—tell you

For the distinction between visibility metrics and website outcomes, see our GEO vs SEO guide.

Before you start

Check the basics:

• You are viewing the correct GA4 property.
• The relevant website is measured.
• The selected date range contains usable data.
• Your reporting timezone is understood.
• Important actions are tracked correctly.
• You know which consent and collection limitations apply.
• Internal testing has not been mistaken for customer activity.

Do not interpret a missing channel row as proof that nobody has ever encountered your brand in an assistant.

It means you need to inspect the available data and its limitations.

Step 1: Open Traffic acquisition

In GA4:

1. Open Reports.
2. Navigate to Acquisition.
3. Select Traffic acquisition.
4. Choose a relevant date range.
5. Review Session default channel group.

Google explains that this report covers acquisition of sessions from both new and returning users.

If you cannot find the report, it may have been removed from the navigation or excluded from the property's report collection. An editor or administrator may need to restore it.

Do not confuse this with User acquisition.

Traffic acquisition helps examine the sources associated with sessions. User acquisition focuses on how users were first acquired.

Those questions can produce different answers for returning visitors.

Step 2: Inspect the AI Assistant channel

Look for AI Assistant under Session default channel group.

Google's current definition says the channel uses a medium of ai-assistant. When a referrer matches Google's recognized assistant list, GA4 sets the medium to ai-assistant and the campaign to (ai-assistant).

Treat that as Google's classification of identifiable traffic, not a complete measure of AI influence.

If the row is absent:

• Check the date range.
• Inspect the underlying source values.
• Review whether collection is functioning.
• Consider whether there are simply no qualifying recorded sessions.

Do not create artificial traffic or change production tagging merely to make the channel appear.

Step 3: Review Session source / medium

Switch the report's primary dimension to Session source / medium.

Inspect the actual source strings in your property.

You may encounter recognizable assistant domains or values, but do not assume every property will contain the same labels.

Search for relevant names such as ChatGPT or Perplexity, then inspect the complete source and medium rather than relying on a broad substring.

Google defines source as where traffic comes from and medium as the type of traffic.

A useful review sheet looks like this:

This is a worksheet, not a set of observed ZocoPilot results.

Why some visits may appear outside AI Assistant

Existing manual tags, classification rules, and the information available for a visit can affect what you see.

For that reason, review both the channel grouping and the underlying source values.

Do not assume that all assistant-associated visits must use referral as their medium.

Do not assume that every recognizable source will necessarily fall into the same channel in every historical reporting period.

Document the values you include in your analysis.

Step 4: Understand where Google's AI features belong

Do not add all Google organic traffic to an “AI traffic” total.

GA4's default channel documentation includes Google's AI Overviews and AI Mode within Organic Search, not AI Assistant.

That classification does not mean your ordinary source/medium report separates every Google visit by whether an AI feature was involved.

Report what the available dimensions actually establish.

A suitable statement might be:

“This report covers identifiable assistant-referred sessions. It does not isolate all visits influenced by Google's AI search features.”

That is clearer than claiming a complete cross-platform AI traffic total.

Step 5: Inspect the landing pages

Source totals tell you where identifiable sessions came from.

Landing pages help explain what people encountered first.

Use an available landing-page report or a compatible Exploration to review the entry pages for your selected session cohort.

Look for:

• Product pages
• Pricing pages
• Documentation
• Integration guides
• Comparison pages
• Educational articles

Choose dimensions with the appropriate scope. A page viewed during a session is not necessarily the session's landing page.

Then inspect the actual page.

Does it help a new visitor understand:

• What the product does?
• Whether it suits their situation?
• Which important restrictions apply?
• Where to go next?

Do not add a sales interruption to every documentation page. Choose the next step that makes sense for the visitor's task.

Step 6: Measure meaningful actions

A visit count is not a business result.

For SaaS, useful actions might include:

• A completed account signup
• A valid demo-request submission
• A completed trial activation
• A purchase
• A documented activation milestone

These are examples, not instructions to implement every event.

Use a definition your team can maintain and verify.

A button click is not always a completed signup. A form submission is not automatically a qualified lead. An account creation is not necessarily an activated customer.

GA4's Traffic acquisition report includes key events and session key event rate. Key events must be configured from the relevant recorded events.

Ask the analytics owner to confirm:

• The event fires at the intended moment.
• Duplicate events are handled appropriately.
• Failed submissions are not counted as successes.
• Test activity is understood.
• Relevant cross-domain journeys are reviewed.
• Collection follows applicable consent requirements.

Do not change consent settings to improve reporting completeness.

Step 7: Use the correct denominator

Dividing seven event occurrences by 80 sessions answers a different question because an event can occur more than once.

Google defines session key event rate as the percentage of sessions in which a key event occurred.

When reporting a specific action, make sure the rate refers to that action, not an unexplained combination of all configured key events.

Also include the counts.

“Six signup sessions from 80 sessions” gives readers more context than “a 7.5% conversion rate” on its own.

Step 8: Report outcomes without overstating attribution

Keep three layers separate.

Visibility

Your brand or page appears in a recorded answer.

Website acquisition

Analytics records an identifiable visit associated with a source.

Customer outcome

A visitor completes a meaningful action, and your business verifies its quality where needed.

These layers can support a broader investigation. They should not be treated as proof of the same event.

A visibility tracker may observe a citation without a click.

GA4 may record a visit without revealing the exact prompt or answer that preceded it.

A CRM may verify that a signup became a customer without establishing every earlier influence.

Do not claim that a particular tracked answer caused a customer visit unless you have evidence connecting those events.

What about Direct traffic?

Some visits arrive without usable source information.

GA4 describes Direct processing in terms of missing referral information or configured source handling.

A buyer might copy a URL, return later, switch devices, or encounter collection limitations.

But Direct also includes other situations.

Do not label all Direct traffic as hidden AI traffic. Do not apply an invented percentage to estimate the missing portion.

If useful, add an optional self-reported attribution question such as:

“How did you first hear about us?”

Treat the response as customer-reported information, not a replacement for session attribution.

Keep the categories distinct when reporting.

Common setup mistakes

Looking only in Referral

GA4 now documents an AI Assistant channel.

Start with the channel, then inspect source and medium for completeness.

Creating a custom channel before reviewing the default

A custom grouping may be useful for a particular reporting need. It is not automatically required.

Google says default channel groups cannot be edited, but custom channel groups can be created.

If you create one, document the rules and why it differs from the default.

Adding assistant domains to unwanted referrals

The unwanted-referrals setting is not a way to create an AI channel.

Google explains that it instructs Analytics to ignore matching referral information.

Do not use it to hide acquisition sources you want to measure.

Adding UTM tags to internal links

Do not manufacture acquisition labels when a visitor moves between pages on your own website.

UTM parameters are intended to identify campaign traffic. Internal navigation should not be relabelled as a new external marketing source.

Sending personal information in event parameters

Do not place email addresses, names, customer secrets or sensitive prompt contents into analytics parameters.

Use an approved measurement plan and the platform's current data policies.

Claiming revenue without working purchase measurement

GA4 revenue reporting depends on correctly implemented commerce events and parameters.

For the purchase event, Google specifically notes the need for value and currency to populate purchase data in Total revenue.

A SaaS subscription recorded elsewhere will not become reliable GA4 revenue data simply because a visitor came from an assistant.

A practical monthly reporting template

Use comparable date ranges and note measurement changes.

If counts are small, say so.

One additional customer can change a small cohort's percentage sharply. Avoid presenting minor percentage changes as a reliable trend.

How to use the findings

Choose a task supported by the data.

Do not rewrite a page merely because a small cohort did not convert.

The visitor may have completed a useful research task without being ready to buy.

For a broader diagnostic process, use our AI visibility audit.

Frequently asked questions

Can GA4 identify every visit influenced by an assistant?

No.

It reports what the collection and attribution information supports. It cannot reconstruct every earlier conversation or missing source.

Where should I look for ChatGPT traffic?

Start with AI Assistant in the session channel report, then inspect the actual Session source / medium values.

Do not assume every ChatGPT-associated visit must be labelled Referral.

Where are Google AI Overview visits?

Google's default channel definition includes AI Overviews and AI Mode in Organic Search, not AI Assistant.

Does GA4 show how often our brand was mentioned?

No.

Website analytics and sampled-answer monitoring answer different questions.

If you need software for the latter, read our AI visibility tool buyer guide.

Does an AI Assistant session prove a recommendation?

No.

A person may have clicked a supporting citation, documentation link, comparison, or another reference.

The visit does not establish the wording or intent of the preceding answer.

Should we expect assistant traffic to convert better?

Do not assume a universal conversion advantage.

Evaluate your own audience, pages, sample size, event definitions and verified outcomes.

Start with the visit you can verify

Open the acquisition report. Inspect the channel and source values. Find the landing pages. Check whether meaningful actions are measured correctly.

Then report the result with its limits.

That gives your team a stronger basis for decisions than combining mentions, clicks and signups into one unexplained visibility score.

Explore ZocoPilot's current features and decide whether they fit the way your team investigates search visibility alongside website performance.