Shopify records where the session behind each order came from. That turns “is ChatGPT sending me anything?” into a question with an exact answer for every order it saw — and a smaller question it genuinely cannot answer, which most write-ups on this skip.
In your Shopify admin, open Analytics → Reports → Sessions by
referrer. Assistant traffic shows up under hosts like
chatgpt.com, perplexity.ai and copilot.microsoft.com.
Set the date range wide; on most stores these are small numbers that vanish inside a
week’s noise.
That report counts sessions. It will not tell you what those sessions bought, which is usually the number you actually wanted.
A hundred sessions that buy nothing and four that spend $600 are the same row in a referrer report. Shopify keeps the customer journey against the order itself — the first and last visit, the referring site, the UTM tags and how many days the whole thing took — so orders can be attributed directly rather than inferred from session totals. That is the number worth reading: not how many people arrived, but what they were worth when they arrived.
A meaningful share of clicks out of an assistant arrive with no referrer attached at all. The link is opened in a way that strips it, or the assistant hides it deliberately. Shopify records that session as direct, and no tool — ours included — can distinguish it from someone typing your domain into the address bar. Anyone who tells you they have solved this is guessing and calling it data.
So the honest shape of the answer is three buckets: orders an assistant is recorded against, orders recorded against something else, and orders with nothing recorded at all. The third bucket is a floor on your assistant traffic, not a measurement of it. Treat a tool that hides that bucket as one that is inflating the first.
Two related cases behave the same way. Answers shown inside Google’s AI Overviews arrive as ordinary Google referrals and cannot be separated out by anybody. Orders placed anywhere but the online store — point of sale, draft orders, an invoice — never have a web session at all.
Attribution tells you what arrived. It says nothing about whether an assistant can read your products well enough to recommend them, which is the upstream half of the same question. We have scanned a few thousand Shopify storefronts the way a crawler sees them, and published what they fall short on: nearly all of them ship a sitemap and a valid robots.txt, because Shopify does that for them, and the gap is almost always the product data itself — whether a page states price, availability and identity in a form a machine can read without guessing.
Free, no signup. Takes about ten seconds.
Shelfglance is the automated version of the above: it reads the journey Shopify already records against each order, attributes it, and reports orders and revenue per assistant, with the unattributed bucket shown rather than buried. The free plan needs no card.
Install Shelfglance on my storeInstalls on any Shopify store. The free plan needs no card, and the dashboard runs on sample data if you would rather look first. Shopify will note that the app has not been reviewed yet: the App Store listing is still in review, and that governs who can find Shelfglance, not who can install it.