Track AI traffic in GA4 through the AI Assistant channel, then inspect Session source / medium and landing pages. To count leads, select your actual lead event and compare it with saved form submissions. Keep Google AI search traffic, missing source data and test submissions separate so your report answers a useful business question: which visits brought enquiries?
A founder deciding whether to fund more AI search content needs more than a rising traffic line. The report needs the source, landing page and action, plus a record that the enquiry reached the business. Those checks can change the next content decision. A page that attracts research visits needs a different response from a page that attracts buyers whose form submissions fail.
Start with the built-in AI Assistant channel
GA4 already has an AI Assistant channel. Google’s analytics release notes describe automatic classification when the referrer matches a recognized assistant: the medium becomes ai-assistant, and the campaign becomes (ai-assistant). You don’t need to create a custom channel just to begin checking recognized AI visits.
Open Reports → Acquisition → Traffic acquisition. Choose Session default channel group, then find AI Assistant. Google’s Traffic acquisition guide explains this report and its source dimensions. If the report doesn’t appear in your navigation, ask someone with editing access to add it.
Use a completed reporting window, such as the previous 28 days. Write down its exact dates and the property timezone. Compare the same number of days, with the same filters, in the previous period. Keep today’s partial data outside that comparison.
Save the sessions total before changing the report. Use that total to check the individual sources later. An empty row means the report found no matching traffic for that view and period. It doesn’t establish that no one discussed your brand in an AI tool.
Our earlier AI Assistant channel guide covers the channel itself. The next step is to check what those visits did after they arrived.
Read the source before changing the grouping
Switch to Session source / medium and inspect the actual values. Keep the source and medium together. A source name alone can hide differences between paid campaigns, tagged links and visits that carry a referrer.
Google defines traffic-source dimensions separately: source identifies the referring platform or place, while medium describes the traffic type. Use those values to understand a row before you decide that its channel needs fixing.
For each apparent AI source, check the landing page and any campaign values. Look for names you recognize, then inspect unfamiliar domains. Don’t add a source to an AI group because its name contains “ai.” A website that reviews AI software may send ordinary referral visits.
Keep a short record of each classification decision. Include the exact source value, the reason you consider it an assistant and the date you checked it. You can then repeat the comparison next month, even if an assistant changes its domain or a campaign adds a source name.
Check paid promotion separately where your source data allows it. A tagged ad visit from an AI platform answers a different spending question from an unpaid click in an answer. If the data cannot separate them, say so in the report instead of calling the whole row earned AI search traffic.

Keep Google AI search in its own view
Google’s default channel definitions place AI Overviews and AI Mode within Organic Search. The AI Assistant channel excludes them. Your AI Assistant total therefore cannot stand for every visit that involved an AI search experience.
Keep two views: assistant traffic and Google organic traffic. Use the second to inspect relevant landing pages without claiming that every Google organic visit came from an AI answer. A standard source row doesn’t establish which part of the search page someone used.
Search Console adds a separate visibility measure. Google’s Search Generative AI performance report announcement describes AI impressions with page, country, device and date views. Those impressions also contribute to the overall performance report. Adding them to the overall total would count some visibility twice.
Use our Search Console AI visibility guide when checking that side of the report. Compare relevant pages and dates, while preserving the name of each metric.
A search impression, a website session and a saved enquiry measure different actions. Keep their counts in separate columns. Avoid a single “AI growth” total that combines them. You couldn’t use that total to decide whether the next task should improve search visibility, the page or the form.
Use session scope for the visit question
Use Session source / medium when you want to know which source brought the visit. Use First user source / medium when you want to study how people first found the site. Google explains these differences in its traffic-source scope guide.
For example, someone may first discover your company through a newsletter, then return through an assistant link. The first discovery and the later visit answer different questions. Keep both when the data supports them, but label them clearly.
Google’s comparison of acquisition reports warns that the reports use different scopes, so their values can differ. Treat a mismatch as a reason to inspect the dimension, dates and attribution settings. Don’t start by assuming that one report lost traffic.
For a practical page review, use the session view consistently. Compare sources, landing pages and the chosen business action within the same date range. Record the exact dimension names beside your saved report so another person can repeat it.
If sales needs a first-touch view as well, keep that as a separate report. Use “source” to mean the same thing throughout each comparison. Switching between first discovery and session source can make the same campaign look stronger or weaker even when customer behavior stays the same.
Find the landing pages that earn useful visits
Break the assistant view down by landing page. If your report won’t show the needed combination, use an exploration with the relevant session dimensions and the same date range. Google’s landing page report defines the landing page as the first page a visitor reaches in a session. For a single assistant, see how to track Perplexity referrals in GA4.
Look for pages that attract repeated visits from relevant sources. Read the page as a new visitor. Does its opening answer the question? Does the next step fit the reader’s need? Can someone understand what the company offers without searching through the navigation?
An informational page can attract useful research visits without producing an immediate enquiry. Check whether it gives readers a clear route to a relevant service, tool or comparison. Keep the link specific. A guide about measuring leads should connect to a tracking check, not send every reader to a general homepage.
Review pages with visits and no lead activity before writing more near-identical articles. Our traffic but no leads checklist covers page and form checks that fit this situation.
Compare pages with similar intent. A service page and a glossary entry have different jobs. Grouping them together can hide a service page problem or make a useful reference page look weak. Record whether each page aims to teach, help someone compare options or collect an enquiry before you judge its results.
Count the lead action you actually need
Select the event that represents a successful enquiry, such as generate_lead, rather than reading every key event as a lead. Google’s key event guidance explains that you choose which events matter to your business. Your property may mark several actions as key events.
Write the lead definition before you calculate a rate. For a service business, we would use a saved, non-test enquiry with the information the team needs to respond. Keep newsletter subscriptions, brochure downloads and audit requests separate when they lead to different follow-up work.
Check when the website sends the event. A button click, an attempted submission and a saved submission represent different stages. Our recommended tracking design sends the success event only after the site confirms that it saved the enquiry. Keep the saved form record as the check on what the event claims.
Then check delivery. A saved enquiry may still fail to reach the CRM or the person handling it. Review those failures separately from acquisition. If a page attracts enquiries but the handoff fails, publishing another page won’t repair that handoff.
When you have permission to test the form, mark the test submission and check its path to the saved record and destination. Exclude it from the business report. Keep personal contact details inside the approved lead system; the marketing report needs counts and outcomes. Our lead tracking cleanup guide covers that wider setup.

Reconcile events with saved enquiries
Compare the selected event count with the saved enquiries for the same period. Check duplicate events, repeat submissions, tests and missing source values. Record unresolved differences instead of forcing both systems to match. A zero in GA4 and a saved enquiry deserve investigation; neither should erase the other.
Consider a hypothetical service business with 80 non-test assistant sessions, six generate_lead events and five saved submissions. The team finds that one submission triggered its event twice. One of the five saved submissions was a marked test. All four remaining submissions carry an assistant source and meet the team’s enquiry definition.
| Measure in this example | Count |
|---|---|
| Assistant sessions | 80 |
| Lead events | 6 |
| Saved submissions | 5 |
| Duplicate events | 1 |
| Test submissions | 1 |
| Valid enquiries | 4 |
The raw event-to-session ratio is 6 ÷ 80, or 7.5%. The valid-enquiry-to-session ratio is 4 ÷ 80, or 5%. That second figure depends on this example’s matching dates and confirmed source records. Call it an enquiry-to-session ratio so nobody mistakes it for a count of unique people who converted.
Use our conversion rate calculator once you have agreed on the numerator and denominator. The arithmetic stays simple. The work lies in deciding which records belong in the calculation.
Keep unknown traffic unknown
Direct traffic cannot tell you how much AI traffic you missed. Google’s campaign and source documentation explains that Analytics processes traffic as direct when referral information is unavailable or an exclusion applies. A rise in Direct alone doesn’t identify the missing source.
Review collection and tagging if Direct rises sharply. Check whether redirects preserve campaign parameters, whether the page collects analytics as expected and whether your own test traffic entered the report. Keep each finding tied to an observed failure.
A custom channel group can help organize confirmed sources that your current view misses. Google’s custom channel group guide describes source rules and their order. Review actual source rows first, then make the narrowest rule that fits the evidence. A broad rule can quietly absorb unrelated visits.
Don’t relabel unknown sessions as AI to make a report look complete. Keep an unknown category and state what you still need to verify. A classification rule can regroup source data that exists. It cannot recover a referrer that the browser never supplied.
GA4 also doesn’t provide a complete list of the questions people asked assistants or every answer that cited you. Use a separate, fixed question sample to study mentions and citations. Keep that sample’s settings and answers so you can compare it later. Website analytics measures the visits you can observe after people click.
Compare changes before expanding the work
Record the page, change and publication date before you judge a result. Review at 7, 14 and 28 days, using those as reporting checkpoints rather than promises of ranking gains. Compare the same pages and measures each time. When traffic stays low, show counts alongside percentages.
The Glasp AEO experiment on GitHub offers a useful research caution. The authors compared changed pages with other pages on the same domain because broader assistant growth could raise both groups. They describe their result as suggestive, with limits from one domain and a short, noisy baseline. We use that work to guide the comparison. It gives us no growth forecast for this site.
Choose similar unchanged pages as a reference group when possible. Record differences in intent and existing traffic. A reference group won’t remove every source of bias, but it gives you more context than a before-and-after screenshot of one page.

If visits rise while valid enquiries stay flat, inspect intent and the next step. If saved enquiries rise while CRM records stay flat, investigate delivery. If the numbers remain too small, keep measuring before making a larger content bet. For help checking that path, start with an analytics and funnel audit.
FAQ
How do I find ChatGPT traffic in GA4?
Start in Traffic acquisition with Session default channel group and find AI Assistant. Then inspect Session source / medium to see the sources behind the total. Check the selected dates and landing pages. A source view measures visits that GA4 can attribute; it doesn’t count every ChatGPT answer that mentions your company.
Does GA4 include Google AI Overviews in AI Assistant?
Google’s default channel definitions keep AI Overviews and AI Mode in Organic Search. The AI Assistant channel excludes those Google Search experiences. Use a separate Google organic view for their broader traffic context and Search Console’s generative AI reports for AI impressions. Keep those impressions separate from website sessions and lead counts.
Do I need a custom channel group for AI traffic?
Start with the built-in channel. Inspect the actual source and medium rows before creating anything new. A custom group may help organize confirmed assistant sources that need a separate reporting view. Keep the rules narrow, check their order and preserve the original source data so you can explain what each group includes.
Can GA4 show the prompt someone used before visiting?
GA4 traffic reports don’t supply a complete record of assistant prompts or cited answers. They show the source information and website activity that your setup collects. Study prompts with a separate, repeatable question sample. Don’t infer a visitor’s exact question from the landing page or treat an observed session as proof of a particular citation.
How do I know whether AI traffic generated a real lead?
Select the event for a successful enquiry, then compare it with the saved form record. Check the source, date, duplicate status and test marker. Confirm that the enquiry reached its destination. Keep events, valid enquiries and sales outcomes separate, and investigate records that lack enough evidence for a source claim.
You're All Set!
We have your request.
Expect a reply from our team within 24 hours. Want to talk sooner? Pick a time below.
Or call us at +1 (914) 297-6638



