Most writing about AI commerce treats it as one thing: a shopper asks ChatGPT, clicks a link and buys. That is the largest path, and it is one of four. We sorted 25,313 AI-attributed orders by how the purchase actually happened. The other three paths are small today, they are growing faster, and none of them looks like a referral.
The paths differ in where the shopper is when they pay, and that decides what a store can observe.
| Path | What happens | What the store sees | Orders |
|---|---|---|---|
| Link click | The assistant links to the store. The shopper clicks and checks out as usual. | A normal session, with a tag on the landing link | 24,011 |
| In-app browser | The store opens inside the assistant's own app and the shopper buys there. | A normal session from a browser that names the assistant's app | 862 |
| Bought inside the assistant | The shopper picks and pays in the conversation. The store is never opened. | An order, and usually nothing else | 434 |
| Agent | An AI agent drives the storefront for the shopper, who confirms the payment. | A session from a browser that declares itself an agent | 6 |
Link clicks are 94.9% of the total. That is why the single-path picture has held up so far, and it is also why it will age badly: every tool built on it measures the one path where the shopper lands on the store with a marker attached.
Between June and October 2026, link-click orders roughly doubled, from 2,364 to 4,696 a month. Orders on the other three paths went from 109 to 396, which is 3.6x. Almost all of that growth is purchases completed inside the assistant.
| Month | In-app browser | Bought inside the assistant | Agent |
|---|---|---|---|
| May 26* | 48 | 0 | 0 |
| Jun 26 | 104 | 5 | 0 |
| Jul 26 | 103 | 5 | 0 |
| Aug 26 | 139 | 33 | 0 |
| Sep 26 | 138 | 81 | 0 |
| Oct 26 | 187 | 208 | 1 |
| Nov 26* | 143 | 102 | 5 |
For scale, here is every path against all orders. AI-attributed orders as a whole went from 5.83 per 10,000 orders in June to 8.5 in October. It is a small channel. It is also one that grew through a period when the stores’ total order volume grew 41%.
| Month | All orders | Link click | In-app browser | Bought inside | Agent | AI per 10,000 |
|---|---|---|---|---|---|---|
| May 26* | 1,678,382 | 1,162 | 48 | 0 | 0 | 7.21 |
| Jun 26 | 4,238,836 | 2,364 | 104 | 5 | 0 | 5.83 |
| Jul 26 | 4,846,033 | 2,835 | 103 | 5 | 0 | 6.07 |
| Aug 26 | 5,101,459 | 4,420 | 139 | 33 | 0 | 9.00 |
| Sep 26 | 6,178,891 | 5,169 | 138 | 81 | 0 | 8.72 |
| Oct 26 | 5,989,102 | 4,696 | 187 | 208 | 1 | 8.50 |
| Nov 26* | 3,653,669 | 3,365 | 143 | 102 | 5 | 9.89 |
The off-site total in Figure 1 is 1,302 orders over the window. Six months ago it would have rounded to nothing.
The assistants do not compete on the same path. ChatGPT sends shoppers out by link. Meta AI completes the purchase itself. Grok and the Gemini app mostly open stores in their own browser. Asking “which assistant sends the most orders” without asking how hides this entirely.
| Assistant | Link click | In-app browser | Bought inside | Agent |
|---|---|---|---|---|
| ChatGPT | 23,051 | 0 | 0 | 0 |
| Gemini and Google AI | 301 | 490 | 47 | 0 |
| Perplexity | 408 | 0 | 0 | 0 |
| Meta AI | 28 | 0 | 377 | 0 |
| Grok | 0 | 330 | 0 | 0 |
| Microsoft Copilot | 139 | 0 | 10 | 0 |
| Claude | 84 | 42 | 0 | 6 |
A practical consequence: a store that sees no Meta AI or Grok in its referral report has not learned that those assistants send it nothing. It has learned that they do not send referrals.
Among orders priced in US dollars, the median order placed inside an assistant was $58.50, against $75.95 for link clicks and $85.70 for in-app browser sessions. One reading is that shoppers hand an assistant the smaller, simpler purchase and go to the store themselves for the larger one. The in-assistant sample is 425 orders, so treat the gap as a direction.
Every figure comes from the order records of Shopify stores running Aimerce, read as one fleet. An order counts as AI-attributed when Shopify’s own record of it names an assistant in one of four ways: the sales channel that created it, a tag on the landing link, the referring site, or a browser that identifies itself as an assistant’s app or agent. Each order is counted once.
How each path is recognised. A purchase inside the assistant is identified by the sales channel Shopify records on the order. A link click is identified by the tag the assistant appends to the landing link or, failing that, the referring site. An in-app browser and an agent are identified by the browser’s own user agent. Where more than one signal is present, the order is assigned to one path in that order of precedence.
What this does not measure. Shoppers who asked an assistant and later came back another way, clicks that arrive with neither a tag nor a referrer, visits from Google’s AI Mode and AI Overviews (they arrive as ordinary Google traffic), and agents that run in the shopper’s own browser without identifying themselves. All four make the numbers here a floor.
Privacy. Only fleet-level aggregates are published. Store counts are rounded to the nearest ten, and no store, brand, product or order is identified anywhere on this page.