AI assistants now send 1 in 80 shoppers — up 5.4x in ten months
There has been a lot of confident arguing about whether AI search is actually sending buyers yet. Most of it has been conducted without evidence, for a reason that is structural rather than lazy: the thing everyone wants to measure is specifically the thing that leaves no trace in analytics.
A shopper asks ChatGPT for a recommendation, reads the answer, and then types the brand name into a browser two days later. No referral is passed. Nothing in Google Analytics, Shopify, or an ads dashboard records that the conversation happened. As far as every tracking system is concerned, that customer arrived from nowhere.
So we asked them instead. The Zigpoll Attribution Index is built from 3,303,402 responses to “how did you hear about us?” questions, collected across 3,237 ecommerce stores and attached to completed orders.
The number
The share of shoppers naming an AI assistant rose from 0.23% in December 2025 to 1.24% in September 2026 — a 5.4x increase in ten months.
| Month | Share of stated discovery |
|---|---|
| December 2025 | 0.23% |
| February 2026 | 0.35% |
| April 2026 | 0.66% |
| June 2026 | 0.70% |
| August 2026 | 1.10% |
| September 2026 | 1.24% |
Growth has been steady rather than spiky. It roughly doubled between December and April, and doubled again by September. The number of individual stores seeing at least one AI-attributed purchase in a month rose from 59 to 237 over the same period, so this is not a handful of outlier merchants moving an average around.
It is still small, and that is the point
At 1.24%, AI assistants account for roughly one stated discovery in eighty. Instagram is at 21.6% and Facebook at 17.5%. Anyone telling you that AI has already reshaped ecommerce discovery is ahead of the evidence.
But no other channel we track moved at a comparable rate. Over the same thirteen months Instagram grew from 17.5% to 21.6%, Facebook from 13.6% to 17.5%, and word of mouth declined from 17.9% to 15.3%. AI is the fastest-moving line on the board by a wide margin, from a small base.
If you are trying to work out when this starts mattering for your budget, the trajectory is a far more useful number than the current level. At the observed rate it is a low-single-digit channel within a year.
A nuance that makes it more credible
AI-attributed purchases turn out to be roughly average in how well they track. 50.3% of them arrived with no usable source recorded, against 48.5% across all channels.
That is worth sitting with, because it cuts against the simple story. AI discovery is not uniquely invisible. It is considerably less invisible than word of mouth (58.7% untracked) or YouTube (60.2%). Some AI traffic does pass a referrer, and some shoppers click through directly from an assistant's answer. The measurement problem here is real but it is not exotic — it is the same referrer-loss problem that has been quietly eroding attribution for a decade, applied to a new surface.
How this was measured
Only responses attached to a completed order are included, so that what a customer said can be compared with the source their store's analytics recorded for that same order. Answers are grouped into channels by an open, rule-based mapping, and answers that cannot be confidently classified are reported separately rather than quietly folded into “other”.
No figure is published for any month with fewer than 50 contributing stores, which is why this series begins in December 2025 even though we have data back to September 2025. Three earlier months exist and are excluded rather than reported thinly.
Stores using Zigpoll are not a random sample of ecommerce, and these figures describe that population rather than the market as a whole. Full tables and method: the Zigpoll Attribution Index.
