Use the same date window in GA4 and Cuebase to record ChatGPT referrals by source. The result is a record of referrals GA4 received, not every visit ChatGPT may have influenced.
Reviewed 13 October 2026.
What this report answers
In Cuebase, Agent Channel Analytics gives you a separately labelled view of AI-referred sessions and order records. Use GA4 alongside it to answer a narrow, useful question: how many sessions did GA4 receive with ChatGPT as the source during this period? GA4 can show a visit as a referral when the browser passes a referrer to your site. It cannot tell you every time someone used ChatGPT before buying. Some assistants, browsers, privacy settings, redirects, and blockers remove or prevent the source information before GA4 receives it.
Google’s guidance on direct traffic says traffic without a clear referral source can appear as (direct) / (none), including traffic affected by missing campaign data, redirects, and ad blockers. Do not add a share of direct traffic to your ChatGPT count. Keep the report honest: it is traffic GA4 recorded, not inferred influence.
Build the check in GA4
Choose a fixed date window
Open Reports → Acquisition → Traffic acquisition. Set a completed window, such as the last 28 full days. Record the property time zone and the exact start and end dates. Avoid today: fresh data can still change.
Use Session source / medium
Set the report’s primary dimension to Session source / medium. This keeps the question about the source of the session, rather than the customer’s first-ever source or an event-level attribution result.
Search for ChatGPT
Use the table search and look for
chatgpt.com / referral. If your historical data contains another ChatGPT host, record that exact value separately. Do not combine look-alike sources just because the names sound related.Record sessions before revenue
Write down Sessions, Engaged sessions, Key events, and Total revenue only if those metrics are available in your report. Then note the denominator you will use. For example, “ChatGPT sessions ÷ all sessions in the same GA4 report and date window.”
For 1–28 October 2026, an example store sees 18 GA4 sessions for chatgpt.com / referral and 12,480 total GA4 sessions in the same property and window. The observed ChatGPT-referral share is 18 ÷ 12,480 = 0.14%. This is not “0.14% of demand came from ChatGPT.” It is the share of sessions GA4 recorded with that referral.
Check the channel before drawing a conclusion
GA4’s default channel group includes an AI Assistant channel for recognized assistant traffic, while Referral remains a separate rule-based channel. A ChatGPT visit may therefore be visible under a source row, an AI Assistant channel, or Referral depending on the data and GA4’s current rules. Start with Session source / medium because it exposes the host and medium. Use the channel group as a second view, not as a substitute for the source row.
If you need a reusable view, create a comparison that uses the same time period and either “Session source / medium contains chatgpt” or an exact source value seen in your property. Keep the filter definition beside the result. A saved exploration with an undocumented regex is hard to audit six weeks later.
Compare it with Cuebase, carefully
In Cuebase, open Agent Channel Analytics, choose the same date range, and look at Sessions by source and Traffic & revenue by source. Cuebase classifies storefront traffic from the web pixel using an explicit UTM marker, a Google AI marker, or a matching referrer host. It shows AI sessions, views, orders, revenue, and a calculated orders-per-session rate by source.
That is a different measurement system from GA4. Cuebase uses its pixel events and Shopify order records for separate session and order attribution. Its interface says attribution is best-effort because assistants can strip referrers. A different number is a prompt to check definitions, consent, date boundaries, and source rules. It is not proof that either tool is wrong.
Keep a small evidence log
Download the worksheet and save one row per review. It asks for the date window, property time zone, source row, GA4 sessions, total GA4 sessions, the denominator, and a note about consent or tracking changes. A small log is more useful than a screenshot with no date range.
Download the ChatGPT referral check worksheet (CSV).
What to do with a non-zero result
First, verify the landing pages and products people reached. Then repeat the same check after a full period. Do not make a sales forecast from a handful of referrals. If you want a store-level view alongside GA4, activate the Cuebase pixel and use the same date window in Agent Channel Analytics. The useful habit is consistency: same question, same window, documented denominator.
Keep the worksheet with the report. Read the parent guide to tracking AI traffic in Shopify, use the shopping-question template when you are defining what to measure, and see the dated backpack study for an example of disclosed search evidence. Then add Cuebase to Shopify when you want a separately labelled store-level view.