Answer engine optimization has become good at measuring answers. Teams can see which prompts name them, how their share of voice compares with competitors, which sources engines cite and how the tone of an answer shifts over time.
What the category rarely does is connect any of that to pipeline. That gap is the missing layer in AEO, and until it is filled honestly, AEO budgets will stay easier to cut than to grow.
AEO measures the answer well, and stops there
AEO measures the answer itself well: visibility, citations, share of voice and sentiment, tracked per engine on a fixed set of prompts. Those are real, repeatable measurements of what buyers are being told.
That is not a small thing. Search reporting trained marketers to look at rankings and traffic; AEO gave them a way to see the content of an answer, the competitors named alongside them and the sources behind it. For diagnosing problems and directing work, the answer layer is the right instrument.
But it stops at the answer. It describes the conversation buyers are having with an assistant, not what they did afterward.
Why does the attribution gap exist?
The gap exists because the influence of an AI answer mostly produces no click, and the data that could show its effect sits in systems the AEO category does not touch.
The channel hides its own effect
An answer that names three tools gives the reader what they needed. Whatever they do next often arrives as a direct visit weeks later, with no referrer and no lineage back to the answer.
The evidence is scattered
AI-referred visits are in analytics and server logs. Self-reported sources are in form tools and the CRM. The richest evidence of all is in sales call notes. None of it sits next to the answer data.
Ownership is split
The person tracking answers is rarely the person who owns the CRM or listens to sales calls. Joining the data means joining teams, and that is slower than buying a dashboard.
So the category has measured what it can see directly. That is a reasonable place to start and a poor place to stop.
The gap is a budget problem
A metric with no visible connection to the business is treated as discretionary, and discretionary spend is the first to go when budgets tighten.
Visibility is a leading indicator. Finance teams accept leading indicators, but only when there is some evidence they lead somewhere. Branded search earned that acceptance over years of watching it move with demand. AEO has not had those years, and a share of voice chart alone does not supply the missing evidence.
The practical consequence: AEO programs get funded as experiments and reviewed as experiments. Without an attribution layer, the review comes down to whether someone senior believes in it.
What does a credible attribution layer look like?
A credible attribution layer reports three layers of evidence together, leans on self-reported data for the influence analytics cannot see, and states its limits plainly.
Three layers, read together
The answer layer (visibility, share of voice and sentiment per engine) is the leading indicator. The visit layer (AI-referred visits, AI crawler fetches of key pages, branded and direct traffic) shows buyers acting on answers. The revenue layer (pipeline where buyers name an assistant, and what comes up on sales calls) shows deals that answers touched. The setup for each is laid out in the AI search attribution guide.
No single layer proves anything. Three layers moving in the same direction over quarters is a pattern that holds up.
Self-reported data at the center
A required “how did you hear about us” field and consistent tagging in sales notes are unglamorous, and they are the only methods that reach influence which never produced a click. When a buyer writes that an assistant recommended you, that is direct evidence no referrer report can provide.
Honesty about limits
Every layer undercounts. Referrers get dropped, self-reported answers are incomplete, and branded traffic grows for many reasons. A credible layer says so in the report itself, treats its counts as floors and describes correlations as correlations.
Modelled revenue claims damage trust
Modelled revenue claims damage trust because every input they rely on is an assumption, and once one assumption is challenged, the audience stops believing the measured numbers too.
The pattern is familiar. Take visibility, multiply by an assumed prompt volume, an assumed click rate and a conversion rate, and present the result as pipeline influenced. It looks like the missing layer. It is actually a substitute for one, and a fragile one. The case against presenting it to finance, and what to present instead, is made in how to report AEO performance to a CFO.
The damage outlasts the meeting. A team caught once with an inflated figure will have its next honest visibility report questioned too. For a category still earning credibility, that cost falls on everyone in it.
The category should build plumbing, not models
The next useful step for AEO is the plumbing that joins answer data with visit data and self-reported pipeline, not a more elaborate revenue model.
Some of that exists today. Server-log analysis can show which AI crawlers read which pages and which visits arrived from assistants. Noma, for example, covers the answer layer and those server-log signals, and does not track conversions, pipeline or revenue. The revenue layer still lives in each company’s own forms, CRM and sales notes, and connecting it is mostly process rather than software.
Any vendor claiming to have fully solved revenue attribution for AI search deserves a direct question: which of your inputs did you observe, and which did you assume?
What to do about it
In rough order of return on effort:
- Make “how did you hear about us” required on high-intent forms, and map it into the CRM. The revenue layer takes quarters to become readable, so start it first.
- Ask sales to tag calls where a buyer mentions an assistant, and which one.
- Segment AI-referred visits in analytics and check server logs for crawler coverage of key pages.
- Keep the answer layer on a fixed prompt set, reported per engine.
- Report the three layers together, as a pattern, with the limits stated in the report.
None of this produces a single revenue number. That is the point. It produces something more durable: evidence a skeptical reader can check.