ChatGPT Referral Traffic in GA4 Without Attribution Drift

OpenAI states that ChatGPT referral URLs include utm_source=chatgpt.com. When the landing request preserves that query parameter and GA4 collects the session, the source can be reported explicitly rather than inferred from a browser referrer. Capture a real landing-page URL before designing filters around third-party lists of possible AI domains.

The parameter describes the source of a click. It does not prove how often the page appeared, which answer text surrounded the link, whether the page was cited in other answers, or why the user chose it. A session is downstream behavior, not a citation impression.

Request the tagged URL and record every hop through www, protocol, locale, authentication, short-link, and trailing-slash normalization. The final page must retain the query string long enough for analytics initialization. A redirect that drops it can move a real ChatGPT visit into referral or direct traffic, depending on what browser information remains.

Also test consent states, ad blockers, client-side navigation, cross-domain transitions, and server-side tagging. GA4 documentation notes that missing campaign or referrer information contributes to direct and none traffic. The absence of a ChatGPT row is therefore not proof of zero visits until the collection path is checked.

Use Session source or Session source and medium when asking which channel initiated a session. First user source answers a different question about the user’s earliest known acquisition. Page location can preserve the landing query string, while the landing-page-plus-query dimension may omit UTM information under documented GA4 behavior. Put the business question beside the selected dimension.

Normalize case and naming in an analysis layer rather than overwriting raw collection. chatgpt.com, ChatGPT.com, and a referrer-derived label can fragment into separate rows. Keep the raw value, normalized provider, channel family, and transformation version so historical reports remain reproducible.

Break sessions down by canonical landing page, page family, device, country, engagement, and a predeclared conversion event. A technical documentation click and a branded profile visit have different intent. Compare AI referrals with similar informational traffic, not automatically with paid search or returning email subscribers.

Small counts need absolute values and long windows. One high-value conversion may matter without supporting a stable rate. A low engagement time may reflect an immediate answer, a measurement issue, or a mismatch. Review the page and event definition before turning the metric into a content judgment.

A complete publisher ledger can show verified OAI-SearchBot requests, sampled ChatGPT citations, tagged referral sessions, and downstream conversions on adjacent rows. Those events may relate, but they are not interchangeable. A bot request does not create a session; a citation may not receive a click; a session can arrive after a copied or shared link.

Report collection gaps explicitly and preserve the observation window. Do not use GA4 to claim total ChatGPT visibility or use Search crawler logs to claim revenue. The credible story is narrower: how many tagged visits were collected, where they landed, what users did, and which technical breaks could cause undercounting.