Track visits from AI search by reviewing website sessions attributed to identifiable AI services, checking the pages those visitors reach, and measuring useful actions afterward. Keep the reporting period, source rules, and attribution settings beside the totals so you can explain what was counted.
This gives you a view of recorded visits. It can't tell you every time an AI answer mentioned your business, or how many people saw a link without opening it. Even an identifiable AI-service visit doesn't establish which prompt or search feature produced the link.
The useful question is specific: which visits can you identify, where did those visitors land, and what happened next?
Start with a consistent traffic report
Keep the date range, reporting unit, and attribution settings with the totals.
In Google Analytics 4, open the Traffic acquisition report and inspect Session source / medium. This dimension describes the source and medium associated with a session. Use the same date range and reporting time zone when comparing periods.
Check the scope of the report before sharing a number. Google's comparison of acquisition reports distinguishes session-based Traffic acquisition from User acquisition, which uses first-user dimensions. A report about how someone originally found the site answers a different question from a report about their sessions.
Attribution also matters. Google documents non-direct last-click attribution for sessions, so a session's assigned source can reflect earlier source information. Avoid describing every session attributed to an AI service as a newly observed click from that service. Keep raw arrival evidence separate when your measurement setup makes it available.
Save the report's name, date range, time zone, dimension, filters, and relevant attribution settings. Those details make the next comparison interpretable.
Review the source evidence before grouping visits
A referrer is information about the site a visitor arrived from, when that information is available. UTM parameters are labels attached to a link for campaign reporting. For example, utm_source names a source and utm_medium names a medium. Google's campaign URL documentation explains how those values reach Analytics and appear in acquisition reporting.
Inspect the values your site actually records. Build a short, reviewed list of AI-service domains from that evidence, and retain the original source values alongside any grouped total. Confirm the exact hostname rather than grouping every source containing a loose fragment such as “ai.”
Treat a campaign tag as a label to examine. A tagged link can be copied or shared, so its label alone doesn't verify the conversation in which a visitor found it. You also can't add tags to every link that another service chooses to display. For links you control, use a consistent naming convention and check the destination after redirects.
Leave unclear sources unresolved. Google describes (direct) / (none) as traffic without a clear referral source, with several possible causes, including missing source information and people accessing a URL directly. A rise in direct traffic doesn't establish that AI search caused the increase.
Google Search needs a separate interpretation too. Google says traffic from AI Overviews and AI Mode is included in Search Console's overall Web search performance data. A Google source label on its own doesn't establish that a visit came from an AI answer.
Interpret a hypothetical week of visits
Keep identifiable arrivals separate from records whose source remains unknown.
Imagine a picture-framing workshop reviewing October 26 through November 1, 2026, in the America/Toronto time zone. The following worksheet is entirely hypothetical. It uses invented arrival records and confirmed quote-request flags to illustrate a reporting decision; it isn't a Google Analytics export or a real business result.
Assume each recorded session appears in exactly one row, based on the arrival evidence captured at its start. Count a session with a confirmed quote request once, even if it contains several form events.
| Arrival evidence in the fictional worksheet | Recorded sessions | Sessions with a confirmed quote request |
|---|---|---|
| Referrer hostname: chatgpt.com | 6 | 2 |
| Referrer hostname: perplexity.ai | 4 | 1 |
| Referrer hostname: google.com | 10 | 1 |
| No referrer or campaign source recorded | 4 | 0 |
The two AI-service rows contain 10 recorded sessions, calculated as 6 + 4. Three of those sessions contain a confirmed quote request, calculated as 2 + 1. Across all four rows, the worksheet contains 24 sessions and four sessions with a confirmed quote request.
Keep the four sessions with missing source information out of the identifiable AI-service subtotal. Their origin remains unknown. Keep the Google row separate too, because the worksheet doesn't establish which Google search experience produced those visits.
A defensible report would read:
In this hypothetical week, 10 of the 24 recorded sessions had an identifiable AI-service referrer. Three of those 10 sessions included a confirmed quote request. Four sessions lacked source information and remain unclassified. These records don't establish the prompts used, whether the workshop was recommended, or the total number of AI mentions.
This wording preserves the difference between an arrival, a useful action, and a recommendation. It also avoids turning a small fictional sample into a benchmark for what another business should expect.
If a GA4 report assigns different source totals from an arrival worksheet, check scope and attribution before trying to combine them. Each view needs its own definition and denominator.
Check the landing page and the action that follows
Review the pages reached by identifiable AI-service traffic. A visit to a care guide may have a different purpose from a visit to a quote page. Look at the content and the next action before deciding that one page is succeeding or failing.
Choose an outcome that matches the page's job. For the fictional framing workshop, a confirmed quote request is useful. A click on the submit button would leave open whether the request was successfully received. Check the event definition and verify the completion path before reporting it as an inquiry.
Keep sessions, event counts, and people distinct. GA4's Traffic acquisition documentation defines sessions, event count, and session key-event rate separately. Several events can occur within one session. If you report a rate, state what the numerator counts and which sessions form its denominator.
Use the findings to improve the page a visitor actually reached. For example, make the scope and inquiry process clearer on a service page, or answer a recurring decision question in a customer FAQ.
Keep referral reporting alongside a separate visibility check
Save the same report definition each week, record tracking changes, and investigate differences before attributing them to content work. A new source rule, a changed event, or a different reporting window can change the comparison you are making.
For visibility inside answers, keep a separate sample of questions, answer dates, mentions, and cited pages. Our guide to measuring AI visibility explains how to define that sample. Referral reporting can show recorded visits and subsequent actions; answer sampling can show what appeared in the particular responses you checked.
For help with SEO and AI search visibility, try HoneyWeRank.