Field notes
AI discovery & ecommerce measurement

How to track AI referrals
and Shopify sales.

Start with identifiable AI sources in your traffic reports. Then check the orders attributed to those visits. Keep a separate record of answers that cite your store: a citation, a visit and a sale are three different events.

The useful business question is whether someone looking for a product reaches a suitable page and buys. A report should help you choose the next improvement, whether that is better buying advice, a clearer product page or a measurement fix.

1. Find the visits you can identify

In Shopify, open Analytics → Reports and find Sessions by referrer. Set a complete date range, then inspect the referrer and available UTM dimensions. Look for the sources actually recorded in your store, rather than assuming all assistant visits arrive under the same label.

Shopify’s acquisition reports cover Online Store visits; they do not establish order counts or sales. Use the separate Total sales by referrer report when investigating attributed sales. Keep bot filtering consistent between comparisons. See Shopify’s acquisition-report documentation.

For ChatGPT, inspect both the referring domain and campaign source. OpenAI documents utm_source=chatgpt.com on outbound search-referral links. Your redirects and tracking setup need to preserve useful source information through the landing journey. See OpenAI’s publisher FAQ.

2. Give known AI sources their own GA4 group

In GA4, review Session source / medium for visits in your reporting period. First-user dimensions answer a different question: where the user was initially acquired. Choose the scope that matches your report. Google explains the traffic-source dimensions.

Create a custom channel group with an “AI assistants” channel. Use a Source condition matching the domains or source tags you have reviewed. Place it above Referral so those visits are captured first, while retaining your intended precedence for paid campaigns. Google documents custom channel groups, including AI assistants.

This is a conservative starter expression for the Source field, not a complete list of every assistant. Review it against your actual source values before applying it:

^(chatgpt|perplexity|claude|gemini|copilot|([a-z0-9-]+\.)*(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com))$

Match the Source field, not an entire URL or a combined “source / medium” string. This expression intentionally avoids matching every domain containing “ai.” Keep unmatched sources visible for review.

Check the rule before using the totals
Source valueExpected source match
chatgpt.comYes
www.perplexity.aiYes
google.comNo
chatgpt.com.example.comNo
email.example.comNo

3. Reconcile the orders before reporting revenue

Use one reporting period, store timezone and currency. Record the attribution model or report definition alongside the figures. Compare the attributed transactions with your order records, checking duplicate purchase events, cancellations and refunds.

A practical reconciliation sheet needs an order identifier, order date, attributed source, sales amount, currency, refund amount and the report used. Keep customer names, email addresses and payment details out of a shareable audit. Record whether the reported amount includes shipping and tax, and whether it is gross or net.

Do not add Shopify-attributed revenue to GA4-attributed revenue: the reports can describe the same orders. Choose a primary commercial report and use the other to investigate differences. “Revenue attributed to AI referrals” is a useful label once the source and order evidence support it.

To decide whether the work pays for itself, examine contribution after product costs, refunds, fulfilment and fees. Attributed sales alone do not show how many orders would have happened without the work. That question needs a separate comparison or experiment.

4. Record citations separately from traffic

Save the exact shopping question, assistant, date, market and answer evidence when checking whether your store is recommended. Record the cited URL, if there is one. Repeat a fixed set of relevant questions over time rather than selecting only answers that mention your business.

A citation may receive no clicks. A referral can arrive without enough information to recover the original question. Keep those gaps visible. For Google AI Overviews and AI Mode, use Search Console’s Generative AI performance report to inspect impressions by page, country, date or device. Google documents a worldwide rollout as of August 31, 2026, while noting that availability can depend on rollout or sufficient impressions. These impressions are included in Web search reporting; do not add the two totals together. They are not a standalone count of visits or sales.

When exporting Google’s AI report, preserve the visible report context too: Google notes that unavailable values shown as “~” or “-” become zeros in the download. A zero in that export may therefore need verification before you call it no visibility.

5. Freeze a baseline before changing pages

Start with a complete 28-day window and a short list of pages you intend to improve. Save the exports, filters and collection date. Keep the same definitions for the next comparison and note promotions, stock changes and releases that could affect the result.

A weekly review that leads to a decision
ObservationNext check
Citations, but no identifiable visitsCheck the links and source collection before estimating a traffic loss.
Relevant visits, little product explorationCheck whether the landing page answers the buying question and links to suitable products.
Product interest, few ordersInspect stock, variants, delivery details and purchase measurement.
Attributed orders, unclear economicsReconcile refunds and contribution before increasing investment.

Download the blank baseline worksheet ↗

The worksheet leaves values empty until measured. Zero means a measured zero; an empty cell means the evidence is missing.

Put the baseline to work

Find the buying question worth improving.

Our Search & AI Discovery Sprint connects a search baseline with practical changes: one buying guide, one collection and up to two product pages, scoped discovery fixes and measurement.

CAD $3,500 · 20 hours over up to four weeks. We agree on the pages and work before payment.

Request your store review ↗

Explore AI search visibility for ecommerce or see our founder-operated work.