What is LLM attribution, and which AI visits can be measured?

In marketing, attribution means splitting the credit for a sale among the channels that made it possible. Attribution for assistants built on an LLM (large language model) applies that idea to a new question: of your store's visits and orders, which ones come from ChatGPT, from Gemini or from Claude? It is not the same as knowing whether an assistant names you, which is a matter of AI citation, nor as raw AI referral traffic: it means joining those visits to what happens next and saying honestly which part cannot be seen. For a store, it is the way to go from "AI recommends us" to "AI brings us orders", or to find out that it doesn't.

How it works

The analytics tool does not know where a visit comes from: it infers it from three signals, and each one has its limit.

The referrer. When someone clicks a link, the browser can send the address of the page they came from in the Referer header. How much it sends is decided by the referrer policy. The one applied by default, if the originating page does not set another, sends only the origin (the domain, without the path) when the jump goes to another site and both use HTTPS, and sends nothing if the destination is less secure. With the no-referrer policy, the header is never sent.⁠^{[1]MDN Web Docs. Referrer-Policy header} So, at best, your store knows the visit came from an assistant's domain, never from which conversation. Google Analytics uses that data: the page referrer field defaults to the browser's value and is used to work out the traffic source.⁠^{[2]Google Analytics, developer documentation. Configuration}

The campaign tags. The utm_ parameters on a link (utm_source, utm_medium, utm_campaign and others) declare the source, the medium and the campaign explicitly, and Google Analytics treats them as the visit's campaign data.⁠^{[2]Google Analytics, developer documentation. Configuration} When a link does not carry them, or a redirect strips them, that information is lost.⁠^{[3]Google Analytics Help. [GA4] Understand (direct) / (none) traffic}

The channel. Since May 13, 2026, Google Analytics has its own channel, AI Assistant, in its default channel group. When the referrer matches a recognized AI assistant, it assigns the visit the medium ai-assistant and the campaign (ai-assistant), and groups it in that channel; Google gives ChatGPT, Gemini and Claude as examples.⁠^{[4]Google Analytics Help. What's new in Google Analytics} The channel's definition also names Deepseek, Copilot and Grok, and explicitly excludes Google's AI Overviews and AI Mode.⁠^{[5]Google Analytics Help. Default channel group} Those two features go to the organic search channel, which according to Google includes the links from AI Overviews and AI Mode.⁠^{[5]Google Analytics Help. Default channel group} And in Search Console, clicking an external link in either one counts as a click in the performance report.⁠^{[6]Search Console Help. What are impressions, position, and clicks?}

What cannot be attributed. A visit with no referrer and no campaign tags ends up as (direct) / (none), the traffic that, in Google's words, has no clear referring source. Among the causes it gives are links without utm_ parameters, redirects that strip them, typing the address by hand and ad blockers.⁠^{[3]Google Analytics Help. [GA4] Understand (direct) / (none) traffic} If someone reads a recommendation in an assistant and then types your store's name into the browser or searches for it on Google, the visit exists, but the credit goes to direct or organic search. And a mention of your brand without a link leaves no signal to attribute.

In a store

A backpack store opens the traffic acquisition report in Google Analytics for September, by default channel group, and breaks the AI Assistant channel down by source and medium. This is a fictional example: the domains and figures are made up to show how to read the report.

Within AI Assistant, each source and medium row has the form example-assistant.com / ai-assistant, and the source says which assistant it came from. In the request the store receives, the only clue is a header like this one, with the origin and no path:

GET /urban-backpack-20 HTTP/1.1
Host: example-store.com
Referer: https://example-assistant.com/

The useful reading is not the AI Assistant row on its own. It is that that row is the minimum: the visits that came from AI Overviews are inside Organic Search, the ones that arrived with no referrer are inside Direct, and the ones that started with a mention without a link are nowhere. If you compare with earlier months, remember the channel has existed only since May 13, 2026.⁠^{[4]Google Analytics Help. What's new in Google Analytics}

How to check it

1. Open traffic acquisition in Google Analytics by default channel group and look for the AI Assistant row, with dates after May 13, 2026.⁠^{[4]Google Analytics Help. What's new in Google Analytics}
2. Break it down by source and medium. The medium will be ai-assistant and the source will tell you which assistant brought each visit.⁠^{[4]Google Analytics Help. What's new in Google Analytics}
3. For Google's AI, go to Search Console. Clicks from AI Overviews and AI Mode are in the performance report, mixed in with classic results,⁠^{[6]Search Console Help. What are impressions, position, and clicks?} and the generative AI performance report gives their impressions.
4. Check which pages get direct traffic. A deep product page that nobody would type by hand and that gets many direct visits is a hint of lost referrers, not proof.⁠^{[3]Google Analytics Help. [GA4] Understand (direct) / (none) traffic}
5. Join the visits to the orders. Clione, with Google Analytics connected, relates the store's orders to the visits that produced them, says which ones arrived from AI assistants and which AI sources cannot be measured, and shows each difference with the margin that says when it may be noise.

Related concept: AI citation

AI citation happens inside the answer: the assistant links your page as a source. Attribution starts afterward, when someone clicks that link and lands on your store. The two are measured with different tools: citations, by reading answers or with the reports that count them; attribution, with your analytics. And they are related imperfectly: many citations may bring no visits, and a mention without a link can end in an order that arrives as direct. That is why a store needs both readings and should not deduce one from the other.

To keep going, the measurement and verification reading path puts the rest of the terms in order, and the entry on assisted conversion explains how credit is split when a sale passes through several channels.