Is AI search sending you customers? How to check in GA4

Someone found your site through an AI answer. Great. Did they visit the pricing page, create an account, or disappear after three seconds? A citation screenshot cannot answer that. Use GA4 to measure identifiable referral visits and the actions that follow. Use a separate research log to measure mentions and citations. These are different datasets, and joining them by guesswork makes both less useful.

Start with the question you want to answer: which sources brought sessions that reached a meaningful outcome? You do not need a custom dashboard on day one. You need a source report, a landing page, and one event that means somebody got value. This guide covers the referral side; our AI mentions and citations guide covers answer sampling.

Start with Session source / medium

Open Traffic acquisition in GA4, set a date range, and change the table's primary dimension to Session source / medium. Search the table for the source you want to investigate. If the report is missing from your sidebar, an editor can restore it through the report library. Google's Traffic acquisition documentation explains the report and its session dimensions.

Look for observed source values such as chatgpt.com or perplexity.ai, rather than assuming every assistant sends an identical referrer. Those are examples to investigate, not a complete or permanent allowlist. Keep the medium beside the source. A manually tagged campaign and an ordinary referral from the same domain can otherwise look identical in your spreadsheet.

Build a small, auditable source list

Export the source rows you actually see. For each candidate, record the exact source value, medium, first observed date, and why you classify it as an AI referral. Keep a separate unclassified bucket. This gives you a report that can be explained to somebody else, instead of a regular expression that silently collects unrelated domains.

If you create a report filter or Exploration, begin with exact matches for confirmed values. Review new rows periodically. Do not include every domain containing the letters 'ai'. A broad filter can count your own newsletter, a directory, or an unrelated campaign. Likewise, a source value is attribution evidence, not proof of an individual person's identity or the exact prompt they used.

Direct traffic is an unknown, not an AI category

Google explains that `(direct) / (none)` means Analytics lacks a clear referral source. Missing campaign parameters and lost referral information are among the possible causes. It does not tell you whether a visitor typed your URL, copied a link from an assistant, or came through another untracked path.

Keep direct traffic separate. If an identifiable AI referral count is small, report that count honestly rather than assigning a percentage of direct traffic to AI. You can add an optional 'How did you hear about us?' question after signup, but keep those answers as self-reported evidence. A customer may have seen you in three places before they registered.

Measure activation before celebrating traffic

Choose one outcome that fits the product. For a project management tool, it might be creating a project and inviting a teammate. For a directory, it might be completing a valid product submission. A page view or scroll is easier to collect, but often says less about whether the visitor found what they needed.

The following numbers are a hypothetical worked example, not RankVyze performance data. Suppose identifiable AI sources bring 40 sessions and three sessions contain your activation event. The observed session activation rate is 3 divided by 40, or 7.5%. If one person fires the event five times in one session, that is still one activated session for this calculation. Event counts and session counts are not interchangeable.

Write the denominator beside the percentage. With only three successful sessions, the rate is fragile: a few additional visitors could change the picture substantially. Compare several consistent reporting periods before changing your budget. Also check whether internal tests, your own browsing, or repeated development events have entered the report.

Use landing pages to choose the next improvement

1. 1

   Find the entry page

   For the same AI source cohort, inspect the landing page. A tutorial and a pricing page serve different intentions; do not average away that distinction.

2. 2

   Check the promise

   Read the page from a new visitor's perspective. Does it answer the likely question, and is the next action understandable without reading your whole site?

3. 3

   Make one useful change

   Add a relevant example, clarify a limitation, or link to the next decision page. Record the change date and avoid redesigning five pages at once.

4. 4

   Compare with context

   Review the next reporting period alongside traffic volume and campaigns. An association after an edit is not proof the edit caused the change.

For links you control, use a consistent campaign naming convention. Do not tag a directory link as 'ChatGPT' just because you hope an assistant will find it. Our UTM guide and UTM builder cover that separate job. RankVyze's AI visibility workflow can complement the report with recorded answer observations; it cannot recover referral data your analytics never collected.