How to track AI traffic for your Shopify store

We checked GA4. We're not getting traffic from ChatGPT.

I hear some version of this from almost every founder I talk to. When we look deeper, the story is usually different.

The traffic is there. It’s just not labeled cleanly, not grouped properly, and not interpreted in context. AI discovery does not behave like classic SEO, so if you expect it to show up like organic search did in 2018, you will miss it.

We start inside Shopify, not GA4

The first step is simple. Open Shopify.

Go to Analytics → Reports and open any existing report.

Instead of scanning traffic sources casually, we go to the Filters section in the bottom right corner, select “Referrer name” or “Order referrer name,” and enter “ChatGPT”. We then repeat the same process for other LLM platforms, for example, Gemini and Perplexity. Shopify's ShopQL also lets you query this directly:

Shopify published their own breakdown of AEO for ecommerce - the Shopify Analytics filter method comes straight from there.

The numbers are rarely dramatic. That is expected. What matters is whether those sessions convert, whether they land directly on product pages, and whether revenue per session looks stronger than your average referral source.

AI traffic is usually more specific. It is often closer to purchase intent than broad organic traffic.

Then we clean the data properly in GA4

We isolate sources like chat.openai.com / referral, copilot.microsoft.com / referral, perplexity.ai / referral, and in some cases bing.com / organic where AI-assisted discovery blends into search.

Then we create a custom channel grouping called AI Traffic. We do not leave it mixed into generic referral.

Once isolated, we compare AI traffic against Organic and Paid using conversion rate, revenue per session, assisted conversions, and engagement time. In most stores we analyze, AI traffic is lower volume but disproportionately high intent.

When it is buried inside referral, it looks statistically insignificant. When grouped correctly, it often tells a different story.

Search Console shows the behavioral shift

After GA4, we open Google Search Console. We are not just looking for rankings. We are looking for patterns.

We filter queries containing words like best, for, without, alternative, under, and vs. Conversational queries have increased steadily, which reflects how people search after interacting with AI systems.

Google is now testing AI-powered dynamic reporting inside Search Console, which is not a cosmetic feature. It signals that traditional static query tables are no longer enough to explain how discovery works in an AI-influenced search environment.

If impressions rise while clicks flatten, that does not automatically mean performance dropped. It can mean that Google’s AI layer is answering more directly inside search. The interpretation matters more than the headline metric.

Bing is no longer optional

Most founders still ignore Bing. That made sense five years ago, it does not make sense now.

Inside Bing Webmaster Tools, Microsoft has introduced an AI Performance report that separates AI-driven impressions and clicks from traditional search data.

Since Microsoft Copilot runs on Bing’s index, this is one of the clearest visibility signals available today. You can see which queries triggered AI experiences and which pages were surfaced inside AI answers.

For the first time, part of AI visibility is being reported directly rather than inferred.

If you are not checking it, you are missing one of the few explicit AI performance layers that exists right now.

AI traffic won't show up as one clean spike labeled "ChatGPT." It shows up as referral sessions, conversational query growth, AI-surfaced impressions, and sometimes branded search lift after a recommendation you never saw happen.

Connect Shopify, GA4, Search Console, and Bing together. You'll start seeing direction instead of noise.

But measurement only matters if your store is actually showing up in AI results. You can check yours in 2 minutes: gpt.atomz.ai

If you don't isolate the signal, you'll assume it doesn't exist.

– Ankit