AI search attribution is the practice of connecting the influence of AI answers to business outcomes: visits, sign-ups, pipeline and revenue. It asks a simple question. When an assistant names your brand to a buyer, what happened next?
The question is simple. The answer, most of the time, is partial, and the tension between those two facts is what this post is about.
What is AI search attribution?
AI search attribution is the work of linking what AI assistants say about your brand to the outcomes your business cares about. It sits between AI visibility, which measures the answers, and revenue reporting, which measures the money.
Traditional search attribution follows a click: a query, a result, a visit, a conversion, with a referrer tying each step to the one before. AI search breaks that chain. The buyer asks a question, reads an answer that names three tools, and forms an opinion. Whatever they do next may happen days later, on a different device, through a different channel.
So AI search attribution is less a single model than a set of signals, each of which captures part of the influence and none of which captures all of it.
Attribution breaks where the click disappears
AI search attribution is hard because most of the influence produces no click, the referrer data that does exist is partial, and the answers themselves change from one run to the next.
Most influence produces no click
An assistant that answers the question fully gives the reader no reason to visit anyone. They read the answer, remember a name, and move on. This is the dynamic behind zero-click search, taken further: the answer is often the whole experience. When that reader later types your brand into a browser, your analytics record a direct visit with no history behind it.
Referrer data is partial
When a click does happen, the referrer is not always passed. Some assistant surfaces send one and some do not, and sessions that start inside an app often arrive with nothing at all. What shows up in an analytics referrer report is a floor, not a total.
Answers vary
The same prompt asked twice can produce different wording, different sources and different brands. There is no stable “ranking” to attach outcomes to, so any link between answers and visits has to be made across many answers over time.
Buying cycles are long
In a considered purchase, the question asked in one month shapes a decision made months later. A short attribution window misses that entirely.
Four methods hold up
Four methods produce evidence that survives scrutiny: AI-referred visits, self-reported attribution, branded and direct traffic trends, and visibility correlated with pipeline over time. Each is incomplete on its own; together they give a credible picture.
AI-referred visits in analytics and server logs
Create a segment in your analytics tool for visits whose referrer is an AI assistant domain. It will undercount, but it is real traffic with a known source, and its trend is informative. Server logs add a second view: they record requests that client-side analytics can miss, including fetches made by AI crawlers reading your pages.
Self-reported attribution
Ask buyers directly. A “how did you hear about us” field on a demo or sign-up form will start to collect answers that name an assistant, and sales call notes capture the same thing in conversation. Self-reported data is imperfect (people forget, and they name whatever they remember last), but it is the only method that reaches influence that produced no click.
Branded search and direct traffic trends
A buyer who learned your name from an answer usually comes back by searching for it or typing it. Branded search volume and direct traffic are where that demand surfaces. Neither is specific to AI, so treat movement in them as supporting evidence rather than proof.
Visibility correlated with pipeline over time
Track your visibility on a fixed set of prompts alongside pipeline, over quarters rather than weeks. If visibility rises for a sustained period and branded demand and self-reported mentions rise with it, that is a pattern worth reporting. It is a correlation, and it should be described as one.
What can AI search attribution not tell you?
It cannot give you a precise, modelled revenue figure for AI search. The inputs such a figure needs are not observable, so any number built from them is an assumption stacked on other assumptions.
Think about what the model would require: how often each prompt is asked, how often an answer leads to a visit, and how often that visit converts. None of those can be measured for AI answers. How to report AEO performance to a CFO covers why a modelled figure loses the argument it was built to win, and what to present instead.
What attribution can do is narrower and more useful: show that AI-referred visits exist and are growing, show that buyers name assistants when asked how they found you, and show whether those signals move with your visibility. That is evidence, not an equation.
Visibility is the indicator, attribution is the evidence
AI visibility measures whether you appear in the answers; AI search attribution looks for what that appearance produced. Visibility is the leading indicator, and attribution is the evidence that the indicator connects to the business.
In practice the two are reported together. Visibility and share of voice explain what buyers are being told. AI-referred visits, self-reported mentions and branded demand show that buyers are acting on it. Neither is complete without the other.
What to do first
In rough order of return on effort:
- Add a “how did you hear about us” field to your highest-intent form. It takes minutes and reaches influence nothing else can.
- Create an analytics segment for visits referred by AI assistant domains.
- Fix a prompt set and start recording visibility, so there is a baseline to correlate against later.
- Agree in advance that you will report a pattern, not a revenue number.
None of these produces a single tidy figure. Together they produce something a skeptical reader can check, which is worth more.