ChatGPT Traffic Analytics: Measure Visits in GA4 & Shopify

Updated 27 October 2026. Originally published 22 January 2026.

ChatGPT traffic analytics can show which visits GA4 attributes to ChatGPT and what happened during those recorded sessions. It cannot count every recommendation, recover every missing referrer or prove that a citation caused a sale. This guide helps Shopify merchants build a useful report while keeping those limits visible.

For other measurement topics, browse our analytics resources. Start here with one question: what do our recorded ChatGPT sessions tell us about visits and outcomes?

Start with GA4’s native AI Assistant channel

GA4 now includes AI Assistant in its default channel group. Google lists ChatGPT, Gemini, DeepSeek, Copilot and Grok as examples. This is a group of sources, so its total is not a ChatGPT total.

Google’s default channel definitions say that a matching AI-assistant referrer sets the medium to ai-assistant and campaign to (ai-assistant). Google AI Overviews and AI Mode are excluded from AI Assistant and are described under Organic Search. Do not treat all Google organic sessions as AI search visits: that channel total does not isolate these experiences.

Use the right scope for your question:

• Session default channel group describes acquisition for sessions, including returning visitors.
• First user default channel group describes how users were first acquired.
• Default channel group, without a prefix, assigns key-event credit using the property’s attribution model.

These scopes answer different questions. Keep session reporting separate from first-user acquisition and event-attribution reports. Google explains the distinction in its traffic-source scope guide.

You do not need a custom channel group to observe recorded AI referrals. Custom groups are optional for a business-specific grouping; document their rules separately from Google’s native classification. Google maintains the native definitions and says they may evolve. Do not assume a new rule has rewritten every historical source/medium row.

Find recorded ChatGPT visits in Traffic acquisition

Follow Google’s Traffic acquisition report guide:

1. Open the correct GA4 property and select Reports > Acquisition > Traffic acquisition. Navigation can vary; if the report is missing, an Editor or Administrator can restore it to the navigation.
2. Choose a completed reporting period and record the property timezone. Keep the same dates for the source breakdown and totals.
3. Use Session default channel group to inspect AI Assistant. Treat this as a starting overview across assistants.
4. Switch the table’s primary dimension to Session source / medium, or add it as a secondary dimension. Search for chatgpt and inspect the actual rows rather than assuming the medium must be referral.
5. Review any historical chat.openai.com rows separately if present. Keep an explicit list of the source labels you include. Do not match every source containing openai without checking what it represents.
6. Save or export the rows and record the report dimensions, filters, dates and included source labels so the next comparison uses the same definition.

For example, a recorded chatgpt.com / ai-assistant row follows the current native medium definition. An older or differently tagged chatgpt.com / referral row is a different recorded label worth reviewing. These examples illustrate labels to inspect, not a promise that either row exists in your property.

A channel label is a classification of recorded traffic-source information. It is not independent verification of the person’s full journey. A broad regex can include unrelated labels; a narrow list can miss a new one. Preserve unfamiliar rows as unknown until you have evidence to classify them.

Direct does not identify hidden ChatGPT traffic

Google describes (direct) / (none) as traffic without a clear referral source. Possible reasons include typing a URL, links in offline documents, missing campaign information, redirects and tracking interference.

A rise in Direct therefore does not establish that ChatGPT sent those visits. Do not allocate an assumed share of Direct to AI, or apply an estimated “lost traffic” multiplier. Investigate your own links and measurement setup before explaining a change.

Likewise, (not set) means GA4 lacks information for a dimension; its cause depends on that dimension. It is not another name for AI traffic. A zero ChatGPT row means no matching sessions were recorded in that report, not proof that nobody discovered your products through ChatGPT.

Keep visits, visibility and automation separate. A product mention or citation does not establish a click. A crawler fetching a page does not establish a human referral or purchase. Source labels alone do not verify that every recorded session was human. Never add bot requests or citation counts to a merchant-session total.

Measure outcomes with clear denominators

First check that your store sends the ecommerce events and values you intend to report. Google’s Traffic acquisition documentation explains that revenue reporting depends on the relevant events and parameters; a missing revenue value is not evidence of no sales.

Build a small scorecard for the same selected ChatGPT sources. Use GA4’s period totals with the appropriate filters; do not add separate source, daily or landing-page tables together to create a total:

Purchases divided by sessions is a count of purchases per session; it is not necessarily the percentage of sessions that purchased. Keep counts next to rates, especially when the sample is small. A handful of purchases can move a rate substantially, and a zero denominator makes a rate unavailable rather than zero.

Compare completed periods of equal length with comparable weekdays. Keep timezone, currency, source inclusion rules and metric definitions fixed. Note promotions, tracking changes and channel-rule changes alongside the comparison. If the earlier value is zero, show the absolute change rather than an undefined percentage increase.

Use landing-page and daily breakdowns to choose what to inspect next: product clarity, stock information, delivery costs or checkout friction. Treat these as investigation leads. A higher rate in a small referral segment does not demonstrate that ChatGPT caused incremental purchases.

Understand ChatGPT shopping and Shopify attribution

Shopify currently describes ChatGPT as a discovery-focused referrer platform, active by default for eligible stores. According to Shopify’s ChatGPT documentation, purchases complete in your online-store checkout, in a ChatGPT in-app browser or a new browser tab for ChatGPT web. Shopify says there is no ChatGPT direct-checkout setting to activate or deactivate.

Catalog eligibility is not a guarantee that your product will be recommended or displayed. Check Shopify’s current requirements for your store; do not interpret catalog access as measured exposure, traffic or orders.

Keep GA4 reporting and Shopify reporting labelled separately. GA4’s session dimensions and event-attribution dimensions can assign credit differently. A purchase-revenue figure attributed to a selected source in GA4 is not a reconciled Shopify sales total or proof of a particular citation-to-sale journey.

Shopify’s analytics discrepancy guidance identifies differences in session definitions, tracking mechanisms, privacy settings, blocking and timezones as reasons its figures can differ from third-party analytics. Shopify’s marketing report documentation explains its selectable attribution models. Check each report’s attribution model, date basis, sales scope, currency and treatment of refunds before comparing outcomes. Do not divide Shopify-attributed orders by GA4 sessions and call the result a GA4 conversion rate.

Shopify also flags a September 2026 session-measurement rollout in that guidance. If your comparison spans a measurement change, annotate the break instead of attributing the whole movement to customer demand.

Keep a repeatable merchant review

Each review should retain the period, included sources, sessions, outcome counts and rates, plus any missing-data warnings. Compare against the same definitions next time. When figures change, check measurement and classification before drawing a commercial conclusion.

Analytics Agent can help review recorded AI-referral sources and comparison periods from connected GA4 data. Its source-label classification differs from Google’s native channel rules; unknown sources remain unknown. Where ecommerce metrics are available, treat the purchase and revenue figures as GA4-attributed outcomes. Availability depends on your connected data and offer, and this is not complete coverage of AI discovery.

Visit our data-use information before connecting a source. You can also keep using GA4 directly: the useful starting point is a consistent recorded-source report, with limits stated beside the numbers.