Between 9 and 14 October 2026, orders attributed to links in ChatGPT answers fell by 63% across the stores we measure, then came back. ChatGPT’s other link type fell by less than half as much over the same days. Most of the drop was not shoppers going away. Those links had stopped carrying the tag that identifies them, and without it a ChatGPT order looks exactly like someone typing the store’s address.
ChatGPT does not pass a referrer on most outbound clicks. What identifies its traffic is a tag it appends to the link. Our main report found that about four in five ChatGPT visits can only be recognised this way. For six days in October that tag was missing from the links in ChatGPT’s answers.
| Day | All orders | Answer-link orders | Product-listing orders | Answer links per 10,000 | Listings per 10,000 |
|---|---|---|---|---|---|
| 2026-09-24 | 175,997 | 108 | 44 | 6.14 | 2.50 |
| 2026-09-25 | 169,776 | 137 | 40 | 8.07 | 2.36 |
| 2026-09-26 | 173,955 | 144 | 41 | 8.28 | 2.36 |
| 2026-09-27 | 173,545 | 147 | 40 | 8.47 | 2.30 |
| 2026-09-28 | 177,200 | 143 | 32 | 8.07 | 1.81 |
| 2026-09-29 | 171,159 | 135 | 35 | 7.89 | 2.04 |
| 2026-09-30 | 195,342 | 133 | 37 | 6.81 | 1.89 |
| 2026-10-01 | 210,723 | 134 | 33 | 6.36 | 1.57 |
| 2026-10-02 | 191,325 | 147 | 59 | 7.68 | 3.08 |
| 2026-10-03 | 187,685 | 132 | 42 | 7.03 | 2.24 |
| 2026-10-04 | 186,546 | 143 | 55 | 7.67 | 2.95 |
| 2026-10-05 | 186,240 | 166 | 45 | 8.91 | 2.42 |
| 2026-10-06 | 180,419 | 152 | 38 | 8.42 | 2.11 |
| 2026-10-07 | 181,902 | 125 | 47 | 6.87 | 2.58 |
| 2026-10-08 | 185,314 | 90 | 49 | 4.86 | 2.64 |
| 2026-10-09 | 181,724 | 53 | 28 | 2.92 | 1.54 |
| 2026-10-10 | 188,393 | 48 | 36 | 2.55 | 1.91 |
| 2026-10-11 | 187,459 | 56 | 41 | 2.99 | 2.19 |
| 2026-10-12 | 185,560 | 59 | 34 | 3.18 | 1.83 |
| 2026-10-13 | 179,815 | 43 | 33 | 2.39 | 1.84 |
| 2026-10-14 | 181,542 | 50 | 23 | 2.75 | 1.27 |
| 2026-10-15 | 193,024 | 70 | 39 | 3.63 | 2.02 |
| 2026-10-16 | 196,543 | 106 | 34 | 5.39 | 1.73 |
| 2026-10-17 | 191,726 | 114 | 27 | 5.95 | 1.41 |
| 2026-10-18 | 198,438 | 110 | 34 | 5.54 | 1.71 |
| 2026-10-19 | 193,514 | 101 | 43 | 5.22 | 2.22 |
| 2026-10-20 | 188,965 | 119 | 39 | 6.30 | 2.06 |
| 2026-10-21 | 185,319 | 89 | 29 | 4.80 | 1.56 |
| 2026-10-22 | 203,324 | 123 | 38 | 6.05 | 1.87 |
| 2026-10-23 | 197,204 | 116 | 33 | 5.88 | 1.67 |
| 2026-10-24 | 196,322 | 109 | 26 | 5.55 | 1.32 |
| 2026-10-25 | 195,272 | 127 | 42 | 6.50 | 2.15 |
| 2026-10-26 | 198,746 | 136 | 52 | 6.84 | 2.62 |
| 2026-10-27 | 200,845 | 110 | 48 | 5.48 | 2.39 |
| 2026-10-28 | 204,139 | 133 | 28 | 6.52 | 1.37 |
| 2026-10-29 | 203,536 | 122 | 38 | 5.99 | 1.87 |
| 2026-10-30 | 222,098 | 123 | 41 | 5.54 | 1.85 |
| 2026-10-31 | 205,440 | 123 | 43 | 5.99 | 2.09 |
In the first week of October answer-link orders ran at 7.54 per 10,000 store orders. From the 9th to the 14th they ran at 2.8. The fall is abrupt on the way down and gradual on the way up, which is the shape of a change rolling out and then being reverted, and not the shape of shoppers losing interest.
ChatGPT sends shoppers to stores through two kinds of link. Orders from product listings went from 2.41 to 1.77 per 10,000 over the same days, down 27%. Orders from answer links went down 63%. Same assistant, same shoppers, same week.
That gap is what separates a measurement failure from a real decline. If people had simply used ChatGPT less for shopping that week, the two series would have fallen together. They did not. Listings did dip, so some part of the week was genuinely softer, or some listing links were affected too. The data cannot say which.
It does bound the damage. If none of the fall was real, the six days produced about 520 fewer attributed orders than they should have. If answer-link demand dipped as much as listings did, the figure is about 300. Either way, several hundred sales happened and were recorded as direct traffic.
For the largest AI shopping channel, attribution depends on a single optional parameter that the assistant adds by choice and can drop without notice. Nothing about the shopper’s browser, the store or the checkout changed during those six days.
Three things follow for anyone reading their own numbers.
A sudden fall in AI traffic is first a question about tagging. Before concluding that demand dropped, check whether direct traffic rose by a similar amount on the same days, and whether a second AI link type fell by the same proportion.
The recovery was partial. In the two weeks after, answer-link orders averaged 5.88 per 10,000, 78% of the earlier rate. Either some links still go out untagged or demand shifted in the meantime. The data cannot tell those apart, and that is the point: once the marker is unreliable, the trend is too.
Comparisons across this window need a footnote. October’s ChatGPT figures are understated in every analytics tool that relies on the tag, including ours. A month-over-month chart that shows a dip in October is showing this incident.
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.
The two link types are separated by the marker ChatGPT adds to product-listing links. The upper estimate of orders counted as direct applies the 1 to 7 October answer-link rate to total order volume on 9 to 14 October and subtracts what was observed. The lower estimate first reduces that expected rate by the decline seen in product-listing orders.
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.