How to Actually See AI Traffic in GA4, Mixpanel, and Shopify (Before Your CMO Asks)

A skincare brand I talked to last month was convinced ChatGPT was driving zero traffic to their site. Their GA4 said so. "Referrals" showed the usual suspects: Google, Facebook, a couple of blogs.

Then we filtered their sessions by hostname and found 1,847 visits from chatgpt.com in the previous 30 days, buried under "Direct" because ChatGPT strips the referrer on most outbound clicks. That was 6% of their organic traffic, invisible.

This is a walkthrough for fixing that in the three tools most ecom and SaaS teams already use. By the end you'll have named segments, custom reports, and a way to correlate what shows up in analytics with your actual AI visibility.

The Hostnames You Need to Filter For

Before anything else, you need the list. AI referral traffic comes from a small set of domains, and most of them are stable enough to hardcode. Here's the working set as of late 2024:

Save this list somewhere. You'll use it in every tool below.

GA4: The Setup That Actually Works

GA4 out of the box will put most AI referrals into "Direct" or "Referral (Other)." Here's the fix.

Step 1: Create a Custom Channel Group

Go to Admin → Data Settings → Channel Groups → Create new channel group. Call it "AI-Aware." Then add a channel above the default "Organic Search" called "AI Assistants" with this condition:

Save it. Now every report that uses channel grouping can be flipped to "AI-Aware" and you'll see AI traffic as its own line item instead of hiding in Direct.

Step 2: Fix the Referrer Loss Problem

Here's where a lot of teams get stuck. ChatGPT often drops the referrer header, meaning sessions arrive with no source at all. To catch those, you need UTM tagging on any link that AI is likely to serve — which sounds impossible, but works in one specific case: your own site's outbound canonical URLs.

If your product page URLs are what AI is citing (they usually are), you can't add UTMs to them retroactively. But you can set up a Google Tag Manager rule that fires a custom event whenever document.referrer contains any of your AI hostnames, and pushes a source=ai_assistant parameter into GA4 as a custom dimension.

The GTM trigger:

Trigger type: Page View
Fire on: document.referrer matches RegEx
Pattern: chatgpt\.com|perplexity\.ai|gemini\.google\.com|claude\.ai

Then a GA4 Event tag pushing ai_referrer as a custom event with the hostname as a parameter. Register that parameter as a custom dimension in GA4 Admin. Now you have a signal even when the source itself gets stripped.

Step 3: Build the Report

Explore → Free-form. Rows: Session source / medium. Values: Sessions, Engaged sessions, Conversions, Revenue. Add a filter: Source matches regex chatgpt|perplexity|gemini|claude|copilot. Save as "AI Referrals — 30d."

You now have a report you can send weekly. The first time you run it against 90 days of data, don't be surprised if AI is already 2-8% of your qualified traffic. The traffic was there the whole time. GA4 was just refusing to name it.

Mixpanel: Custom Properties Are Your Only Move

Mixpanel doesn't have channel groups. It has properties, and you can build reports on top of them. The setup is cleaner in some ways than GA4 because you're not fighting a taxonomy.

Step 1: Capture the Referrer as a Super Property

In your Mixpanel initialization code, capture the initial referrer and parse it:

const aiHosts = ['chatgpt.com', 'chat.openai.com', 'perplexity.ai', 
                 'gemini.google.com', 'claude.ai', 'copilot.microsoft.com'];
const ref = document.referrer;
const aiSource = aiHosts.find(h => ref.includes(h)) || null;
mixpanel.register({   'ai_referrer': aiSource,   'ai_referred': !!aiSource,   'referrer_full': ref });

Now every event a user fires carries ai_referrer (e.g., chatgpt.com) and a boolean ai_referred. These become super properties, meaning they attach to every subsequent event in the session without you having to think about it again.

Step 2: Identify the Funnel

Build a funnel: Landing → Product View → Add to Cart → Purchase. Break down by ai_referrer. A DTC coffee brand we work with ran this exact funnel and found that Perplexity-referred visitors converted at 4.2% while ChatGPT-referred visitors converted at 1.8%. Same brand, same site, wildly different intent.

Step 3: Cohort the AI Users

Create a cohort: "Users where ai_referred = true in last 90 days." Then compare their LTV, repeat purchase rate, and engagement against your baseline cohort. If AI-referred users are stickier or worse, you now know whether to double down or stop caring.

Shopify: Where It Gets Weird

Shopify's built-in analytics is the worst of the three for this job. It doesn't expose a referrer filter deep enough to isolate AI sources, and its "Sessions by referrer" report groups things unhelpfully. You have three practical paths.

Here's the SQL shape once you land it in a warehouse:

SELECT 
  CASE 
    WHEN referring_site LIKE '%chatgpt.com%' THEN 'ChatGPT'
    WHEN referring_site LIKE '%perplexity.ai%' THEN 'Perplexity'
    WHEN referring_site LIKE '%gemini.google.com%' THEN 'Gemini'
    ELSE 'Other'
  END AS ai_source,
  COUNT() AS orders,
  SUM(total_price) AS revenue,
  AVG(total_price) AS aov
FROM shopify_orders
WHERE created_at > CURRENT_DATE - 90
  AND referring_site IS NOT NULL
GROUP BY 1;

What "Good" Looks Like

Rough benchmarks from what we see across ecom brands with intentional AEO efforts, running for at least 60 days:

One Thing to Automate Before You're Done

Set up a weekly Slack or email digest that dumps: AI referrals last 7 days, breakdown by source, conversion rate, revenue attributed. It takes 20 minutes with GA4's scheduled reports or a Zapier flow off Mixpanel. Do it before you forget.

The teams that treat AI referral tracking as a "we'll build a dashboard someday" project never build the dashboard. The teams that get a weekly number in Slack start noticing week-over-week changes and asking why. That's the point.

Where Foxish Fits

Foxish tracks the visibility side (how often AI recommends you across ChatGPT, Claude, Gemini, and Perplexity) and pairs it with the traffic side (what those recommendations actually drive in GA4). The Traffic Impact feature does the correlation we described above out of the box, so you don't have to run the visibility-vs-clicks analysis by hand every month. If you've got GA4 wired up and want to see whether your visibility gains are showing up as actual sessions, start a free scan →. First scan is free, no card needed.