How to track AI referral traffic to your Shopify store

Where AI traffic actually shows up

The mechanism is simple. When a shopper clicks a link inside an AI assistant, the browser usually sends a referrer with the request, and that referrer is a domain you can recognize: chatgpt.com, perplexity.ai and the equivalents for other assistants. Anything that can read the referrer on a storefront page view can classify that session as AI-referred.

On Shopify, the supported way for an app to observe storefront events is a Web Pixel. Pixels run in a sandboxed context and receive events like page views and checkouts, which means an app can record the referrer on arrival and follow that session through to a completed order. That gives you traffic and conversion broken down by assistant, rather than a lump of unattributed sessions.

If you are not running anything purpose-built yet, you can still get a first look. In whatever analytics you already use, find the referrer or source dimension and filter for assistant domains. It is manual, and it will miss things, but it tells you within an hour whether this channel exists for your store at all.

Now the caveat that shapes everything else in this guide: what you capture this way is a floor. It is the subset of AI-influenced buying that happened to leave a referral trace. The real influence is larger, and nobody can tell you by how much.

A floor is still worth measuring properly, because this one is small, moving fast and unusually valuable per session. Adobe Analytics data, reported by TechCrunch in April 2026, put AI traffic to US retail sites up 393% in Q1 2026 against the same quarter a year earlier, and found that by March 2026 AI traffic converted 42% better than non-AI traffic — a reversal from March 2025, when the same comparison had it converting 38% worse (Adobe Analytics data, reported by TechCrunch, April 2026). That analysis covers over 1 trillion visits to US retail sites alongside a survey of more than 5,000 US respondents. None of which tells you the size of the unmeasured portion — it tells you the measured portion is worth pulling out of Direct.

How each source tends to appear

Assistants behave differently, and the way each one lands in your reports is worth knowing before you start reading numbers.

The fourth row is the one that quietly distorts everything else. Google's AI Overviews are now found on 20%+ of all searches, and where they appear they reduce click-through by nearly 60% (SparkToro, June 2026). None of that separates itself out in your referral data. It arrives inside ordinary organic search and looks like an ordinary decline.

Why AI influence is under-counted

This is the part most tracking guides skip, and it is the part that matters most. A large share of AI-influenced buying never appears as an AI referral at all. There are several separate reasons, and they stack.

The assistant answers without a click. This is the biggest one. An assistant can name your product, describe the material, quote a price and summarize your return policy inside its answer. The shopper reads it, decides, and never visits a page. The influence is total. The referral is zero.

The shopper takes a detour. Someone reads a recommendation, then opens a new tab and searches your brand name, or types your domain directly. That session lands in Organic Search or Direct. The assistant did the work. Another channel gets the credit.

The referrer never arrives. Some assistant surfaces do not pass a referrer. Referrer policies, in-app browsers, privacy modes and link handling in mobile apps all drop or truncate the data before it reaches your pixel. This is not something you can configure your way out of.

The purchase happens elsewhere. In-chat checkout means an order can complete without a storefront session existing at all. There is no page view to attribute.

The first of those four is the only one anyone has managed to measure from the outside, and the numbers are stark. Pew Research Center, working from the actual browsing activity of 900 US adults in March 2025, found that users clicked a traditional result link on 8% of visits where an AI summary appeared, versus 15% where none did — and that just 1% of visits to a page carrying an AI summary produced a click on a cited source (Pew Research Center, July 2025). The trend line has kept moving in the same direction since: SparkToro's analysis of a Similarweb clickstream panel covering US desktop and mobile searches from January to April 2026 put the zero-click rate at 68.01% of Google searches, up from 60.45% in 2024 (SparkToro, June 2026).

Put those together and a simple illustration makes the shape clear. Suppose an assistant recommends your product to 100 shoppers. If 30 click through, 25 search your brand instead, 20 type your domain, and 25 buy or bounce without ever leaving the chat, your analytics shows 30 AI-referred sessions out of 100 influenced shoppers. Those numbers are invented for the sake of the arithmetic. The point is not the ratio, it is that the measured slice and the influenced population are different sizes, and you cannot see the gap from inside your analytics.

A practical setup, end to end

1. Capture the referrer on every session

   Install a Web Pixel or configure your analytics to record the full referrer on storefront page views, and to carry that source through to the order. If you can only see the referrer on the landing page and not at checkout, you get traffic but not conversion, which is half the answer.

2. Build an explicit AI source list

   Maintain a list of assistant domains you classify as AI, and keep it in one place you can update. New surfaces appear regularly. A source list you wrote once and forgot will quietly under-report as the landscape shifts.

3. Split assistants apart, not into one bucket

   One combined "AI" row hides the useful signal. Break it out by assistant so you can see which ones send buyers rather than browsers. Conversion rate by source is more actionable than total AI sessions.

4. Baseline Direct and branded organic before you change anything

   Because a lot of AI influence surfaces as Direct or brand search, you need a before picture. Record the current level of both, then watch how they move as your assistant visibility changes. Movement there is soft evidence, not proof, but it is the only view you get of the invisible portion.

5. Widen your attribution window

   AI discovery often happens well before the purchase. A shopper asks an assistant on Sunday, thinks about it, buys on Thursday after clicking an email. Last-click attribution gives all the credit to the email. If your reporting only looks at last click, AI will look smaller than it is regardless of how well you capture referrers.

6. Track visibility on the same schedule

   Run a fixed set of real shopper prompts against the assistants you care about, on a repeating schedule, and record whether you get mentioned. Read that trend next to your referral numbers. Without it you cannot tell a tracking gap from an absence of recommendations.

Cuebase covers the two measurement halves directly: Agent Channel Analytics uses a Shopify Web Pixel to capture AI-referred traffic and conversion broken down by AI source, and Share of Voice probes ChatGPT, Perplexity, Gemini, Claude and Google AI with real shopper prompts to measure how often you are mentioned versus competitors.

Why you need visibility and traffic together

Referral traffic and share of voice answer different questions, and either one alone will mislead you.

Referral traffic tells you what converted. It is downstream, concrete and tied to revenue. But it cannot distinguish between "assistants never recommend us" and "assistants recommend us constantly and nobody clicks". Those two situations produce the same flat line in your reports and demand completely different responses.

Share of voice tells you whether you are being recommended at all. It is upstream and it moves first. Visibility can rise well before clicks do, because being named in an answer does not require anyone to click. If you only watch referral traffic, you will conclude that catalog work did nothing during exactly the period when it started working.

• Visibility flat, traffic flat. You are not being recommended. The work is upstream: product data, attributes, crawler access.
• Visibility rising, traffic flat. You are being named but not clicked, or the clicks are arriving without referrers. Expect this to be common.
• Visibility rising, traffic rising. The channel is working and you can start reading conversion by source seriously.
• Visibility flat, traffic rising. Check your source list first. Something new is probably being classified as AI, or an old surface changed its referrer.

The third case is the one that repays a separate row in your reporting, because when AI sessions do arrive they tend not to behave like the rest of your traffic. In the same Adobe Analytics data reported by TechCrunch in April 2026, AI traffic to US retail sites showed 37% higher revenue per visit, 48% longer time on site, 13% more pages per visit and a 12% higher engagement rate than non-AI traffic (Adobe Analytics data, reported by TechCrunch, April 2026). Averaged into a general Direct bucket, all of that disappears. It is only visible if you kept the sources apart.

Reading the numbers without fooling yourself

A few habits keep this honest. Read trends over weeks, not days, because assistant output varies between runs and small counts swing wildly. Compare like with like, since a change in your source list changes the number without anything real happening. And when AI-referred sessions grow, check whether Direct grew too, because the two often move together and that correlation is one of the few hints you get about the unmeasured portion.

The goal is not a precise attribution model. It does not exist for this channel yet. The goal is a defensible floor that trends in a direction, plus a visibility measure that tells you why it is moving. That is enough to decide where to spend effort, which is the only thing measurement is actually for.

FAQ

Sources

Every external figure quoted above is linked to its original publisher with the date attached, so you can check it and see how current it is. Note what each one actually measures: all three cover Google search or aggregate web analytics, and none of them measures click-through from inside an assistant conversation, which is precisely the gap this guide keeps pointing at.

1. Pew Research Center. Google users are less likely to click on links when an AI summary appears in the results. 22 July 2025.
2. Fishkin, R. In 2026, Less than One Third of Google Searches Still Send a Click. SparkToro, 9 June 2026.
3. Adobe Analytics data, reported by TechCrunch. AI traffic to US retailers rose 393% in Q1, and it's boosting their revenue too. 16 April 2026.