How to Track AI Referral Traffic and Leads in GA4

To track AI referral traffic in GA4, open Traffic acquisition, find the AI Assistant channel, and break it down by Session source / medium. Select your demo or signup key event to see which visits produced an enquiry. For Australian SaaS reporting, add a Country equals Australia comparison and check lead quality in your CRM.

The setup changed in 2026. Google introduced the AI Assistant channel on 13 May, so a custom regex is no longer the starting point for everyone.

This guide covers the current workflow, the gaps worth checking, and a reporting template you can use without buying another analytics platform.

Research checked against Google's documentation on 7 October 2026.

What counts as AI referral traffic in GA4?

AI referral traffic means visits that follow links from AI assistants. The useful distinction is between a human clicking a recommendation and a crawler fetching your content. A crawler request in your server logs is not a buyer session.

According to Google's default channel definitions, recognised assistant referrers receive the ai-assistant medium and belong to AI Assistant. Examples include ChatGPT, Gemini, Copilot and Grok.

Google's AI Overviews and AI Mode belong to Organic Search, not AI Assistant. Putting every AI-related visit in one bucket hides that difference.

Keep those categories separate before comparing results. Our AI referral traffic business case explains why the channel deserves attention; this guide focuses on measuring your own outcomes.

Step 1: Find AI traffic in Traffic acquisition

Use the report that answers the question you are actually asking. Google distinguishes Traffic acquisition from User acquisition: the first uses session-level dimensions; the second describes how users were first acquired.

Start with Traffic acquisition when reviewing AI-driven visits during a reporting period.

1. Open your GA4 property and go to Reports.
2. Find Acquisition, then Traffic acquisition. Navigation can differ by report collection; use report search if needed.
3. Set a consistent date range, such as the last 28 complete days.
4. Select Session default channel group as the table dimension.
5. Find AI Assistant, then add Session source / medium as a secondary dimension.

Record sessions, engagement and your selected lead event. If the channel is empty, inspect source rows before concluding that AI sends nobody.

Avoid filtering to referral medium alone. That excludes recognised visits assigned ai-assistant. Also avoid mixing First user source with Session source in the same comparison: they answer different acquisition questions.

Step 2: Check sources the default channel misses

The built-in channel is your baseline. A custom group is useful when you need a broader, explicitly maintained source list or a consistent view of older recorded referrals.

Inspect Session source / medium for domains you recognise. Check actual rows before copying a list from a tutorial. Domain names change, and a broad match for ai can catch unrelated websites.

For a supplementary group, follow Google's custom channel group workflow:

1. Go to Admin, then Data display, then Channel groups.
2. Create a group based on your existing channel definitions.
3. Add a channel named AI sources audit.
4. Match the recognised ai-assistant medium, or match Source against your verified assistant domains.
5. Place this channel above Referral and other rules that could claim those visits.

For a Source matches regex condition, this is a starter list to validate, not an exhaustive platform directory:

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

Use it on Source, not the combined Source / medium field. Keep the native and audit views side by side; differences are something to investigate. Custom grouping cannot recover an origin that was never recorded.

Step 3: Track enquiries that actually completed

An AI traffic report becomes useful when it connects visits to a business action. For SaaS, that could be a submitted demo request, completed trial signup or confirmed sales booking.

Choose one primary outcome first. A button click tells you someone tried to start an action; it does not prove the form succeeded.

Implement an event that fires after confirmed completion, such as generate_lead for a successful enquiry. If you use a custom name like demo_request, document its trigger so everyone reports the same action.

Then go to Admin > Data display > Events and mark the appropriate event as a key event. Google's key event instructions explain the setup and note that marking an event does not change historical data.

Test one successful submission and one failed submission. The successful path should fire once; the failed path should not count as a lead. Use Google's DebugView guide with debug mode enabled to inspect the events.

Check external booking tools too. A click out to your scheduler is not a confirmed appointment. Validate the completion integration before giving that click a pipeline value.

Step 4: Build an Australian SaaS reporting view

Apply Country equals Australia through a report comparison or filter. Use the same date range and event definition for AI Assistant and Organic Search, then review the landing pages that attract each audience.

This separates Australian visitors from a global traffic total. It does not tell you whether their company buys in Australia. Use the CRM's company and qualification fields to establish that.

For wider market context, our Australian AI search statistics collect adoption research. Your acquisition report should still use your own visitors and enquiries as its baseline.

Build a saved report or Exploration with session source, landing page and country. Track sessions, the selected key event and its session key event rate. Google's metric definitions distinguish session rate from event counts.

Do not divide every recorded key event by sessions and call it a session conversion rate. One session can trigger an event more than once.

Here is an illustrative reporting template, not client performance data:

The native and custom AI groups can overlap. Never add their totals together. In this example, six demo sessions justify reviewing the enquiries, not declaring that AI always converts better.

Step 5: Connect AI enquiries to pipeline in AUD

GA4 records actions. Your CRM tells you whether those actions became qualified opportunities or paying customers. Join the reporting process without treating every form submission as revenue.

Use a consistent qualification rule: relevant company, suitable buyer, real problem and a plausible buying process. Keep duplicate contacts and existing customers out of new-lead totals.

For an illustrative Australian SaaS calculation, assume three qualified AI enquiries, an expected annual contract value of AUD $6,000 and a historical qualified-lead win rate of 20%:

Expected annual contract value from this cohort
= 3 qualified leads x AUD $6,000 x 20%
= AUD $3,600

That is expected contract value, not booked revenue or profit. Replace the assumptions with your CRM's actual cohort results as deals progress. Subtract delivery costs and acquisition costs before discussing profitability.

Also separate attribution from causation. A prospect may encounter your brand through Google, a colleague and ChatGPT before booking. A recorded AI touchpoint does not prove the sale would disappear without it.

For the commercial context around programme spend, see our GEO agency pricing guide for Australia.

How should you measure AI Overviews and AI Mode?

Use Search Console for Google's AI search visibility and GA4 for behaviour after arrival. Do not relabel all Google organic visits as AI traffic.

Google announced dedicated Search Generative AI performance reports on 3 June 2026, with worldwide rollout noted on 31 August. These provide a separate visibility view for generative AI features; that data also remains within overall Search performance.

Compare relevant pages and periods with their GA4 landing-page outcomes. Keep country and device filters aligned. Google's guide to combining Search Console and Analytics explains why clicks and sessions differ.

Treat this as complementary reporting, not a perfect join that identifies every AI-generated visit or enquiry.

What will this setup still miss?

GA4 measures recorded website activity. It does not reveal every prompt, recommendation or citation that happened before a click. Missing source information also limits what a channel rule can identify.

Add an optional question to your signup flow: How did you first hear about us? Offer AI assistants as a choice and let people name the tool. Keep those answers separate from observed referrals because people can misremember their journey.

Review technical gaps before making budget decisions:

• Are internal tests appearing in acquisition reports?
• Does the lead event fire twice after a reload?
• Does an external booking journey preserve useful measurement?
• Are missing tags or consent choices limiting recorded activity?

Review sources monthly, record configuration changes and compare complete periods. Small cohorts need patience: one deal can move the result sharply. The practical goal is a repeatable baseline you can trust.

Turn the report into a next action

Start with one Australian reporting view, one completed lead event and one monthly CRM review. Then ask which landing pages earn qualified conversations and which attract visits that go nowhere.

At Avinash Vagh, we help SaaS founders connect AI visibility work to a measurable acquisition plan. If you want to identify the gaps behind your report, book an Australian GEO scoping call.