AI Traffic and Attribution: How to Measure What ChatGPT, Gemini and AI Mode Send Your Store

The short answer: you can measure most of what ChatGPT, Gemini, Copilot and Perplexity send your store, and you can't measure Google's AI Mode separately at all. In GA4, visits from recognized assistants land in the AI Assistant channel. Visits from assistants Google doesn't recognize stay in Referral. AI Overviews and AI Mode count as Organic Search. Any AI visit that arrives without a referrer lands in Direct. Shopify adds its own view of orders from AI channels, Search Console counts AI Mode clicks inside ordinary search data, and server logs show what AI agents fetch. Attribution is the job of putting those pieces in one report without counting anything twice.

This guide is for Shopify and DTC merchants, and the agencies who report for them, who have been asked "how much revenue does AI bring us?" and found no report that answers it. Every platform rule here comes from the vendor's own documentation, listed under Sources. Where a vendor doesn't document something, we say so.

AI traffic attribution

The practice of identifying visits and orders that started with an AI assistant or AI search feature, such as ChatGPT, Gemini, Copilot, Perplexity, Google AI Overviews or AI Mode, and assigning them revenue credit under a stated attribution model. It covers both the classification (which sessions count as AI) and the credit rule (how much of an order the AI touch earns).

Where does AI traffic show up in your analytics?

In four places, and each one gets there by a different mechanism. GA4 assigns every session to a channel by checking its source, medium and campaign against a set of rules, and an AI visit only carries what the AI product and the shopper's browser pass along. Whether a click carries a referrer depends on the AI product, the link and the shopper's browser, so the same assistant can show up in two channels. A product found in Google's AI Mode arrives from google.com and looks like any other search click.

Two of those buckets can be measured. Two can only be estimated. Bad AI reporting either presents the first two as the whole story or invents a precise number for the other two.

What does GA4's AI Assistant channel count?

Google added an AI Assistant channel to GA4's default channel group in 2026. Google's "What's new" page dates the release to May 13, 2026. Google's channel definitions describe it as the channel by which users arrive from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok. The rule has two parts: when the referrer matches Google's list of AI assistants, GA4 sets the medium to ai-assistant and the campaign to (ai-assistant), and the channel then catches any session whose medium exactly matches ai-assistant.

AI Assistant channel

A channel in GA4's default channel group for sessions that arrive from AI assistants. GA4 assigns it automatically when the referrer matches Google's list of recognized assistants, setting the medium to ai-assistant and the campaign to (ai-assistant). Google's AI Overviews and AI Mode are excluded and counted in Organic Search.

Three limits follow. Google doesn't publish the full list of assistants. The rule needs a referrer, so visits without one are invisible to it. And Google's "What's new" entry doesn't say the change applies to historical data, so year-over-year growth in the channel is partly just the channel appearing. Google's June 11, 2026 Source Group dimension, which groups AI sources such as ChatGPT and Perplexity, is one Google does say is populated retroactively.

Because the channel matches on medium, it also catches any link you tag yourself with utm_medium=ai-assistant, which is handy for links you control inside an AI product and easy to misuse, so keep a list. The full rule set and a fix for each gap are in GA4's AI Assistant channel: what it counts and what it misses.

Which AI visits does GA4 miss, and how do you catch them?

The ones without a recognized referrer. In practice that means three groups: AI tools that aren't on Google's list, AI visits whose referrer was removed on the way, and Google's own AI search features, which GA4 counts as search on purpose.

AI tools that aren't on Google's list

These still arrive with a referrer, so they show up in Referral under their own source name. You can catch them with a custom channel group: a copy of the default group with an extra channel whose condition matches AI sources by regex. Google's help page says custom channel groups can be applied to reports retroactively, which makes them the only way to build a consistent AI trend line that starts before May 2026.

The regex we use, and the one our report template ships with, is applied to the source with GA4's "matches regex" condition:

.*(chatgpt|openai|perplexity|gemini\.google|bard\.google|copilot|claude\.ai|deepseek|grok|meta\.ai|chat\.mistral|you\.com|phind|poe\.com).*

GA4's regex conditions require a full match, which is why the pattern starts and ends with .*. Two warnings. It over-matches: any source containing one of those words counts, including a site that merely has "copilot" in its name, so scan the matched sources monthly. And the list goes stale as assistants launch and domains change, so check it against your Referral sources every quarter.

AI visits with a stripped referrer

A referrer is a header the browser sends, and the sending page can withhold it. Under the browser default, strict-origin-when-cross-origin, a cross-site click sends only the origin, such as https://chatgpt.com/, never the conversation. A page or link marked no-referrer or rel="noreferrer" sends nothing at all. Those visits look exactly like someone typing your URL, so GA4 files them under Direct. The AI Assistant channel can't see them, and neither can a regex.

Referrer stripping

When a visit arrives without the HTTP Referer header that names the page the shopper came from, because the source page, the link or the app withheld it. Analytics tools then classify the visit as Direct. It is the main reason AI traffic is undercounted.

You can't recover those visits afterwards. You can watch for the pattern: Direct sessions landing on deep product pages, rising alongside AI referrals, suggest AI traffic is hiding there. Microsoft's Clarity documentation gives the same warning for its own AI channels. Report it as a caveat, not a number.

Google AI Overviews and AI Mode

Google's channel definitions are explicit: the AI Assistant channel excludes AI Overviews and AI Mode, and Organic Search includes them. A click from either arrives from Google search like any other, so no channel group can split it out. The only partial view is Search Console, covered below.

What can each measurement tool see?

No single tool covers AI traffic end to end. Each sees a different slice.

GA4

GA4 is where most stores build the report, because it puts AI sessions, key events and revenue beside every other channel. The Traffic acquisition report (Reports → Acquisition → Traffic acquisition) defaults to Session default channel group, so the AI Assistant channel appears there as a row. Add Session source / medium as a secondary dimension to see which assistant sent what. Google notes that for revenue to appear, the purchase event has to carry both value and currency parameters.

Shopify Analytics

Shopify sees what GA4 can't: orders placed through AI channels whether or not the shopper's visit was tracked by a script. The agentic storefronts page (Sales channels → Agentic) reports Sales, Orders, Online store sessions and Online store conversion for each AI channel. Shopify's help page says sales there are an aggregate of referral-based sales and sales through direct checkout, and that Shop app sessions, referrals and sales are included. It also says agentic storefront analytics aren't yet available for headless stores.

Outside the Agentic page, the Sessions by referrer report shows referring channel, platform and medium for every session, and the marketing reports offer five attribution models (more on those below). How to track ChatGPT traffic and revenue on Shopify walks through both views and how to reconcile them with GA4.

Search Console

Search Console sees AI Mode, but only mixed in. Google's Search Central documentation says AI feature traffic is included in overall search traffic, in the Performance report under the Web search type. The Performance report's help page sets the rules: a click on an external link in an AI Overview or AI Mode counts as a click, standard impression rules apply, an AI Overview occupies a single position, and an AI Mode follow-up question counts as a new query. Neither page describes a filter that separates AI traffic. Google's separate Generative AI performance report, announced June 3, 2026 and available to all sites from August 31, 2026, comes closest: it shows impressions in AI Overviews and AI Mode, combined, by page, country, device and date, with no clicks and no queries. AI Mode traffic hides in Organic Search covers the estimation methods that exist and how wide their error bars are.

Microsoft Clarity

Clarity, Microsoft's behavior-analytics tool, has two AI channels of its own: AIPlatform for organic links in an AI chat answer and PaidAIPlatform for paid ads inside one. Microsoft lists Claude, Copilot, Gemini, ChatGPT and Perplexity as supported, counted only from their standalone sites, and says data from before the feature's release isn't reclassified. Use it to cross-check GA4's counts.

Server logs

Logs answer a different question: are AI products reading your pages? OpenAI's crawler docs separates OAI-SearchBot (which surfaces sites in ChatGPT search), GPTBot (training), OAI-AdsBot (ad landing-page checks) and ChatGPT-User, which visits a page when a ChatGPT user's question triggers it. Perplexity documents the same split between PerplexityBot and Perplexity-User. A rise in user-triggered fetches on a product page suggests shoppers are asking about it in AI chats, even without clicks. It isn't traffic or revenue, so use it as a leading signal only.

How do Shopify's AI channel orders fit with GA4?

They overlap, and they won't match. Shopify's agentic storefronts page says orders from AI channels display in the admin with channel or referrer attribution. On ChatGPT, Shopify sends the shopper to your own checkout, in ChatGPT's in-app browser or a new tab, so GA4 can see that visit if its tag fires and a referrer survives. On Copilot, Google AI Mode and Gemini, Shopify-powered direct checkout can take the order inside the chat when you've turned it on, and there's no storefront session for GA4 to record.

So GA4 tells you how AI-referred sessions behave on your site, and Shopify's Agentic page tells you how many orders AI channels produced, including ones that never touched your site. Report both, labelled, and never add them: one ChatGPT-referred order can appear in both.

Which attribution model should you use for AI traffic?

Classification decides which sessions count as AI. The attribution model decides how much credit they get when an order follows, and it can move AI revenue more than any channel setting.

Attribution model

The rule an analytics tool uses to divide credit for a conversion among the touchpoints that preceded it. Last-click models give all credit to the final qualifying visit. Data-driven models spread credit based on how each touchpoint changed the observed likelihood of converting.

GA4 offers three models in its attribution reports: data-driven, paid and organic last click, and Google paid channels last click. The first-click, linear, time decay and position-based models were removed in November 2023. All of them exclude direct visits from credit unless the whole path was direct, so a shopper who clicks from ChatGPT on Monday and types your URL to buy on Thursday can still credit ChatGPT, inside the lookback window.

The lookback window defaults to 90 days for most key events (30 or 60 are options) and 30 days for acquisition events. The reporting attribution model, data-driven by default, changes event-scoped dimensions such as Source for historical and future data, but not session-scoped ones such as Session source. Those use paid and organic last click, and a session started by direct entrance takes the user's existing UTM values. Both live under Admin → Data display → Events → Attribution settings.

Shopify's marketing reports offer five models: last non-direct click (the default), last click, first click, any click and linear. Last non-direct click excludes direct visits unless the journey was only direct. Last click, first click and linear include them.

Our recommendation: one last-click number as the headline, because it's conservative and easy to defend to finance, and one assist view beside it, such as Shopify's first click or linear, to show journeys AI starts and other channels finish. Name the model on every chart.

Does AI traffic convert better than other channels?

Sometimes, and the published studies disagree on how much. Google says that when people click from search results pages with AI Overviews, the clicks are higher quality and the visitors more likely to spend more time on the site. That's Google's own claim, without a number. Third-party studies vary by sample, period and how they defined AI traffic, which is why their headline figures conflict. Does AI-referred traffic convert? compares the studies side by side, with who ran each one, when and on what sample.

For your own store, compare AI conversion rate with your site-wide rate over the same period, on enough sessions to mean something: forty AI sessions and two orders is an anecdote, not a rate. Check whether AI visitors land on product pages or blog posts, because the landing-page mix moves the rate as much as the source does.

How do paid AI placements change the picture?

Once you buy ads inside an AI product, paid and organic clicks from it have to stay apart. Clarity already splits them into AIPlatform and PaidAIPlatform. In GA4, tag every ad landing URL you control with your own campaign parameters, then check in Session source / medium that paid clicks aren't sitting in an AI row meant for organic answers.

What OpenAI itself provides for measuring ChatGPT ads is a separate question, covered in ChatGPT ads conversion tracking on Shopify. What the ads product sells and how its auction works is in ChatGPT Ads for ecommerce.

What should an AI measurement plan include?

A measurement plan says which question each number answers, where it comes from and what it can't see. Write it before the dashboard, or the dashboard will answer the easy questions instead of the ones you were asked.

Two numbers need care. AI share of sessions can rise just because other channels fell. AI revenue per session is the most useful for a budget conversation, because it folds conversion rate and order value into one figure you can set against paid channels. If you want to size the opportunity before building anything, the AI revenue calculator gives a directional estimate from your own inputs.

Key event

GA4's name for an event you mark as important to your business, such as a purchase. Attribution reports and the Key events metric are built on them. For a store's revenue report, the purchase event is the one that matters.

How do you set up AI traffic tracking in GA4?

No developer needed. Paths are from Google's help pages.

1. 01

   Check what GA4 already catches

   Open Reports → Acquisition → Traffic acquisition, find the AI Assistant row, and add Session source / medium as a secondary dimension. Then scan the Referral rows for AI tools: anything there is a source Google's list doesn't cover yet.

2. 02

   Create a custom channel group

   In Admin, under Data display, open Channel groups and click Create new channel group, which starts from a copy of the default group. Click Add new channel, name it (for example "AI assistants, all"), click + Add condition group, set the source condition to matches regex with the pattern above, and click Save channel.

3. 03

   Put the channel in the right order

   Click Reorder, drag your AI channel above Referral (and above the built-in AI Assistant channel if you want a single AI row), click Apply, then save the group. GA4 assigns traffic to the first channel that matches.

4. 04

   Confirm your attribution settings

   In Admin → Data display → Events → Attribution settings, note the reporting attribution model and the key event lookback window. Write both into your report so every reader knows which rules the AI revenue follows.

5. 05

   Link Search Console

   In Admin, under Product links, open Search Console Links, click Link and choose your property and web data stream. GA4 then gets the Google organic search queries and traffic reports, with data arriving 48 hours after collection.

6. 06

   Annotate the dates that break trends

   Footnote May 13, 2026 (GA4's AI Assistant channel appears) and September 21 to 23, 2026 (Shopify's session measurement change) in your report. Comparisons across either date need it.

The regex, the full rule set and the reasons behind each choice are in the GA4 AI Assistant channel guide. If you report in Google's dashboard tool, note that Google renamed Looker Studio back to Data Studio in April 2026. Google's documentation says reports built on its GA4 connector are subject to Google Analytics Data API quotas and can't use GA4 segments or comparisons.

What does a monthly AI reporting rhythm look like?

Weekly for checks, monthly for the report, quarterly for the rules. AI traffic is small and noisy for most stores, so weekly revenue mostly reports noise.

• Weekly (15 minutes). Scan Referral for AI sources your regex misses, and check the AI channel didn't drop to zero, which usually means a broken tag rather than a collapse.
• Monthly (one report). AI sessions, AI share of sessions, AI key events, AI purchase revenue, AI conversion rate against the site and AI revenue per session, each beside last month and last year. Add Shopify's Agentic sales as a separate line and the Search Console Organic trend with its AI Mode caveat.
• Quarterly (an hour). Re-check the regex, re-read Google's channel definitions for rule changes, and confirm the attribution settings haven't moved.

The monthly report has two readers: a marketer choosing which assistant to work on, and a finance lead who wants a number that won't move when a setting changes. The monthly AI-attributed revenue report walks through each section, how to read month-over-month moves and what to say to a CFO. If you're also measuring an on-site AI agent, measuring agentic commerce ROI covers that separate stack, and LTV math for AI-acquired customers covers what those customers are worth after the first order.

What should you do with the numbers?

Start the series now, because the baseline is the hard part and a store that starts later will be comparing against nothing. Then let the numbers point at the work. Rising AI sessions with flat AI conversion point at what visitors find when they land: thin product pages, missing sizes, unclear shipping. Flat AI sessions point upstream, at whether assistants can read and recommend your products at all. The AI shopping channels guide maps how products get into ChatGPT, Copilot, Gemini and Perplexity in the first place.

If the numbers show AI visitors arriving and leaving without buying, or not arriving at all, Commerce GEO is where we work on the product data and content that decide whether assistants recommend your store.