Attributing leads to AI searches: AI traffic attribution

AI traffic attribution is hard because assistants like ChatGPT and Perplexity often send no recognizable source on a click, so that visitor shows up in your analytics as “direct” instead of as an AI referral. The result: you do get leads from AI searches, but your report does not show it. In this article you will read why that happens, how to still recognize traffic from ChatGPT and Perplexity, and how to tie a concrete lead back to an AI mention without kidding yourself.

What is GEO attribution?

GEO attribution is assigning visitors, leads and revenue to AI searches, that is, to moments when a language model named or linked to your brand. Where classic attribution ties a conversion to a Google click or an ad, GEO attribution tries to do the same for answers from ChatGPT, Perplexity, Google AI overviews and comparable systems.

That is fundamentally harder, because an AI answer is usually “zero-click”: the user gets the answer inside the conversation itself and only visits your site later, or not at all. The mention does its job (your brand gets considered) without a measurable click to match. GEO attribution is therefore less about recounting visitors and more about reconstructing a first touchpoint that has become invisible. How you even end up in those answers we cover in the GEO guide; this article is purely about measuring it.

Why does AI traffic disappear into your “direct” numbers?

AI traffic disappears into your “direct” bucket because the referring source (the referrer) is not passed to your website on many AI clicks. Analytics tools like GA4 need that referrer to tie a visit to a channel. If no source comes along, the system sees no difference between someone typing your URL and someone clicking a link in an AI answer: both become “direct”.

On top of that, many AI answers live inside an app or an embedded browser. A user who reads your name in ChatGPT, remembers your brand and only goes directly to your site the next day was in a sense driven by AI, but leaves no digital trace of that origin. The result is a systematic underestimation: your AI traffic is almost always bigger than your dashboard shows. That is precisely why AI visibility deserves its own yardstick alongside your regular analytics, a point we develop further in the five key metrics of AI visibility.

How do you recognize traffic from ChatGPT and Perplexity in GA4?

Part of your AI traffic is recognizable: when the source comes along, traffic from ChatGPT shows up with domains like chatgpt.com and from Perplexity as perplexity.ai in your referral report. The problem is that GA4 places those visitors among your regular referrals by default, where they get lost in the noise. So you have to isolate them yourself.

The practical approach consists of three steps:

1. Create a dedicated channel group. Define a custom channel group “AI” in GA4 that filters sources on the known AI domains (among others chatgpt.com, perplexity.ai and the assistant domains of the major providers). Place that group above “Referral”, so AI traffic does not get absorbed there first. Keep the list of domains up to date, because new ones keep appearing.
2. Filter on the source parameter. Since mid-2025, ChatGPT attaches a recognizable source parameter (like utm_source=chatgpt.com) to a portion of its outbound links. That helps, but it does not cover everything, and the behavior differs per assistant and per type of answer. So do not rely on it blindly.
3. Accept that this is only the visible top. Even with a perfect channel group you only measure the clicks with a source. The zero-click mentions and the “direct” visitors stay out of view. This number is a floor, not a total.

If you want to understand more deeply how you even end up in the answers of these platforms, read getting found in ChatGPT and getting found in Perplexity. Measuring only makes sense when there is something to measure.

Which signals still tie a lead to AI?

The most reliable way to tie a lead to AI is to let the visitor say it themselves. Because the technical origin is so often missing, “self-reported attribution” is not a last resort for AI traffic but your primary source.

Concretely, that works like this:

• Add a “How did you find us?” question to your most important forms. An open or multiple-choice field with options like “via ChatGPT or another AI”, “via Google”, “recommendation” and “already knew you”. This captures exactly the leads your analytics writes off as “direct”.
• Ask it in the first sales conversation. B2B deals run through people. A simple question from your sales team (“where did you first come across us?”) often yields the answer “I asked ChatGPT”, and that belongs in your CRM.
• Watch the content of the request. Someone who comes in with a surprisingly well-phrased, comparative question (“do you also do X for sector Y, because that is how it was described to me?”) has often been primed by an AI answer. It is not hard proof, but a signal worth noting.

None of these signals is watertight on its own. Together they give a far more honest picture than your analytics alone. Combine them: cross the self-reported answers with your recognizable AI traffic and with your AI visibility, so that a rise in mentions, referrals and reported origin confirm one another.

Make it easy on yourself to collect those answers, too. Keep the origin field short and concrete, because the more choices you give, the sloppier people fill it in. Make sure the answer flows automatically into your CRM, so sales and marketing look at the same number. And agree internally on who definitively records the origin of a won deal; without that agreement, the data dilutes within a quarter. This is not an analytics project but a process matter, and that is precisely what makes it feasible for a small team.

How do you measure AI attribution when traffic stays invisible?

If the click stays invisible, you shift your yardstick from traffic to visibility and to the lead itself. Instead of asking “how many visitors did ChatGPT send?”, you ask “am I named in the answers my buyers get, and is the number of leads that name AI as a source growing?”.

You do that with two measurement layers side by side. The first is visibility measurement: structurally test how often and how correctly the major models name your brand on the prompts your audience really asks. Here, specialized AI visibility tools help to track those mentions across models and over time. The second layer is the lead side: the percentage of new leads that give AI as their origin, and how those leads convert relative to other channels.

The CI approach here is deliberately sober. We do not steer on a vanity number like “AI traffic”, because that number is by definition incomplete and easy to make look nicer than it is. We steer on the combination of growing, correct mentions and qualified leads that name AI. Promising that we can attribute every AI lead exactly would be dishonest; the honest promise is that you map the trend reliably and can steer on it. For B2B, where the buying journey is long and runs across multiple touchpoints, that trend-based approach is more realistic than a seemingly precise click attribution.

What does this mean for your B2B reporting?

It means your reporting has to treat AI as a separate channel with its own honest measurement logic. Put three numbers side by side: your recognizable AI referrals from the custom channel group, your AI visibility on core prompts, and the share of leads that give AI as a source in your form or CRM. None of the three is complete, but together they tell a consistent story that is much closer to reality than a single number.

It is important to adjust the expectation level internally. As long as a good chunk of AI traffic comes in as “direct”, a low AI number in your analytics is not proof that AI does not work. It is often the opposite: proof that you are not measuring it well yet. Whoever understands this stops steering on the wrong knob and starts investing in visibility and in recording origin at the source.

The short summary

AI traffic attribution fails by default because ChatGPT and Perplexity often send no source, which makes leads disappear into your “direct” numbers. You restore the view with a dedicated GA4 channel group for AI domains, with UTM and source filters where they exist, and above all with self-reported attribution: a “how did you find us” question in your forms and your sales conversations. Combine that with structural visibility measurement, and steer on qualified leads instead of on an incomplete traffic number.

Do you want to map your AI leads reliably and let your GEO visibility grow alongside your reporting? Explore our generative engine optimization (GEO) or plan your free intake.