Most AI visibility reporting stops at the point where it gets interesting. The brand is mentioned in seven prompts, up from two. Everyone nods, and nobody can say what it was worth.
That is a problem, and not only for the marketing team. A metric that cannot be connected to revenue eventually gets cut, however good it looks on a slide. Here is the chain that connects the two, and the specific places it breaks.
The five stages
Visibility
Whether a generative engine names you when someone asks a buying question in your category.
What to measure: Mention rate across a fixed prompt set, citation count, recommendation position, share of answer against competitors, and the sentiment or context of the mention.
The critical step: Record a baseline before any work starts. Without it, every subsequent number is a claim rather than a result.
Sessions
Visits arriving from an AI surface, plus visits influenced by one.
What to measure: AI referral sessions separated from generic referral traffic through a custom channel grouping, alongside branded search volume as a proxy for influenced but unattributed arrivals.
Leads
Enquiries you can trace, at least partly, to AI-led discovery.
What to measure: Form submissions from AI referral sessions, plus responses to a self-reported attribution field on your primary form.
Qualified opportunities
Leads your sales team accepts as real.
What to measure: Qualification rate for AI-sourced leads against your other channels. This is the stage that tells you whether the traffic is genuinely better or just differently counted.
Pipeline and revenue
Value created, attributed and influenced.
What to measure: Pipeline value and closed revenue from AI-sourced opportunities, reported separately from influenced revenue so the two are never confused.
Where the chain breaks
Three failures account for almost every broken loop we see.
- No baseline. The most common and the most damaging. Visibility work started before anyone recorded the starting position, so the result cannot be proven even when it is real.
- The dark funnel. Zero-click queries rose from 56 percent to 69 percent after AI Overviews rolled out. A large share of AI-influenced discovery produces no referral at all. The person reads the answer, does not click, and arrives later through a branded search. Your analytics credits are direct or organic, and the AI answer that caused it never appears.
- No channel separation. AI referrers sitting inside generic referral traffic, indistinguishable from a link in someone's newsletter. As of late 2025 only around 16 percent of brands were measuring AI search performance systematically at all.
You will under-count AI-sourced revenue. Put that in the board deck yourself, with your reasoning, before someone else discovers it and treats the whole number as unreliable.
The cheapest fix on the list
One optional field on your primary form asking how the person found you. It costs a day to implement and it catches a meaningful share of the discovery that analytics structurally cannot see.
It is imperfect. People misremember, and some skip it. It is still the only mechanism most companies have for observing the dark funnel at all, and imperfect observation beats none.
What good reporting looks like
- Visibility metrics reviewed monthly, against a recorded baseline, on a prompt set that does not change without being noted.
- AI referral sessions and branded search volume reported side by side, so movement in one can be read against the other.
- AI-sourced leads and their qualification rate reported alongside every other channel, using the same definitions.
- Attributed and influenced revenue reported separately and labelled clearly.
- A stated confidence level on the whole thing, so the number is understood as directional rather than exact.
Why leading indicators still earn their place
None of this makes visibility metrics worthless while the revenue is still small. Leading indicators exist precisely because they move first.
Citation position and prompt coverage change months before session volume does, and session volume changes before pipeline. If you wait for the revenue line to justify the work, you will start the work two years after the companies who did not wait, and by then the positions worth holding will be held by someone else.
Measure the leading indicator honestly. Connect it to the chain above. And be the first person in the room to say how much of it you cannot yet see.
Get your AI visibility score
We will run your brand against a tracked prompt set, record a baseline, and show you how visibility connects to demand in your category.