One retailer. Two reports.
A closer look at how Shopify recorded referrals and attributed orders for one unnamed auto-parts retailer. Five views, with the dates, denominators and reporting definitions alongside the figures.
Report A is the Shopify session extract. Report B is the channel-performance export. They use different reporting definitions, and each is presented separately.
The session extract.
1 July to 30 September 2026, pulled 4 October. Orders and sessions from the Shopify session extract used in OptiVis's original referral study.
Session report: 19 orders from 151 ChatGPT sessions, against 76 from 3,988 Google search sessions.
| Channel | Orders | Sessions | Orders per 100 sessions |
|---|---|---|---|
| ChatGPT | 19 | 151 | 12.58% |
| Google search | 76 | 3,988 | 1.91% |
- ChatGPT is the smaller sample here. Google search sent about 26 times as many sessions.
- Report A counts differently from Report B, the channel export. The two are shown apart and never added.
The channel-performance export.
The July, August and September monthly exports, added. This is a separate reporting extract: its 129 ChatGPT-attributed orders and 1,519 sessions cannot be added to, or substituted for, Report A's 19 orders and 151 sessions.
Channel export: ChatGPT converted at 8.49%, with an unclassified Google row close behind at 7.80%.
| Channel | Orders | Sessions | Conversion |
|---|---|---|---|
| ChatGPT | 129 | 1,519 | 8.49% |
| Google unknown | 70 | 897 | 7.80% |
| Google organic | 33 | 3,221 | 1.02% |
| Direct | 39 | 89,465 | 0.04% |
- Google unknown is a Google row the export types as unknown, not organic or paid. Leaving it out would flatter every comparison, so it stays in.
- Direct has 89,465 sessions, mostly non-buying traffic. It is not a baseline, and it still produced 39 real orders.
The full quarter, channel by channel.
| Channel | Sessions | Orders | Sales | Conv. | Avg. order |
|---|---|---|---|---|---|
| ChatGPT | 1,519 | 129 | $10,804.33 | 8.49% | $83.75 |
| Google unknown | 897 | 70 | $6,402.17 | 7.80% | $91.46 |
| Google organic | 3,221 | 33 | $3,307.11 | 1.02% | $100.22 |
| Direct | 89,465 | 39 | $3,158.37 | 0.04% | $80.98 |
- ChatGPT includes 230 paid sessions with no orders (August 49, September 181). Without them the rate would read 10.01%, so they stay in.
- Average order is sales divided by orders. ChatGPT’s is below both Google rows.
Inside the quarter, orders attributed to ChatGPT peaked in August.
September check, same exportGoogle unknown: 37 orders from 401 sessions, $4,280.26, 9.23%. Ahead of ChatGPT on sales and conversion that month.
| Month | Orders | Sessions | Conversion | Sales | Paid sessions (0 orders) |
|---|---|---|---|---|---|
| July | 23 | 266 | 8.65% | $2,172.80 | 0 |
| August | 60 | 539 | 11.13% | $4,901.79 | 49 |
| September | 46 | 714 | 6.44% | $3,729.74 | 181 |
- Paid ChatGPT sessions are inside these sums and produced no orders: August 49, September 181.
Two complete periods.
Calendar 2025 beside April through September 2026. The chart compares the average attributed orders per month in each window.
In the export’s own history, orders attributed to ChatGPT averaged 3.00 a month in 2025 and 25.67 a month from April to September 2026.
| Period | Months | Orders | Average orders a month | Sessions | Sales | Conversion |
|---|---|---|---|---|---|---|
| Calendar 2025 | 12 | 36 | 3.00 | 729 | $2,851.35 | 4.94% |
| Apr to Sep 2026 | 6 | 154 | 25.67 | 1,998 | $12,615.45 | 7.71% |
- Each column is one period’s average, not a month. The export records the change; it does not say what caused it.
Read each report on its own terms.
Report A uses ShopifyQL session and sales extracts. Report B uses the Sessions, Orders and Sales columns in Shopify's channel-performance exports. They cover the same retailer, but use different reporting definitions. The available files do not establish why the extracts differ. We keep both labeled and do not combine them.
Quarter figures use three full, non-overlapping monthly files. The 2026 historical window uses six full monthly files, April through September; calendar 2025 uses its annual export. Trailing-period exports that overlap those months are excluded. Conversion is orders divided by sessions. Period averages are orders divided by the number of months in the stated window.
ChatGPT includes the chatgpt and chatgpt.com rows and their paid sessions. Google organic and Google unknown remain separate because the export labels them separately; unknown does not establish paid traffic. Spend is absent from these exports. The figures describe attributed orders, not additional orders caused by OptiVis, a controlled incrementality test or a guarantee of future results.
The retailer remains unnamed. Customer records and raw store exports are private. On a review call, we can walk through the source definitions and the limits of these comparisons.
Start with your own baseline.
We can review how your store is recorded today and scope the work from there.