Why the old numbers stop working
A ranking was a position you held. You could watch it move from 42 to 22 to 10 and know you were on the right track. An AI answer is assembled fresh each time, from searches you do not see, and the sources it cites change constantly. Run the same question twice and you can get two different sets of names. Run it once and conclude you are invisible, and you may simply have been unlucky.
The click is also gone for much of the value. When an assistant recommends a firm, the person may search the firm’s name later, or ring, or do nothing for a month. Analytics sees a brand search or a direct visit and cannot connect it to the answer that caused it. Wil Reynolds of Seer Interactive put it as “the box just got way blacker”; we would add that it is also possible to be visible and anti-recommended, named in an answer only so the reader can be told to look elsewhere. Visibility on its own is not the number. See Ranked but Not Recommended.
The four measures
1. Momentum: a share, not a position
We agree ten prompts with the client, the questions their buyers ask, in the buyers’ words. Each month we run every prompt a set number of times on the platforms that matter to that client and record whether the client is named, what it is named for, and in what company. The result is a share: named in 7 of 20 runs for this question, 0 of 20 for that one.
This is a momentum measure. It tells you whether the work is on the right track and which questions are close. It is not a headline KPI and we will not present it as one, because a share of answers is not a share of customers.
2. Mentions from sources the model already trusts
Fan-outs increasingly search named publications and named people directly. If a model goes to a trade publication for an answer and that publication has never mentioned you, you were never in the running. So we track mentions of the client, trended, from the sources that show up in fan-outs for its sector: trade press, professional bodies, the named commentators in that field, and the review platforms buyers use. Rising is good; flat is a finding; the list of sources is itself a deliverable.
3. Referrals from people, not algorithms
We split referral traffic into two piles: sources where a person chose to share the link (LinkedIn, NotebookLM, newsletters, named blogs, forums) and sources where an algorithm placed it. The first pile is the only evidence a site’s content is liked by the humans whose publications the models trust. It is also the pile that the “ten times more content at half the cost” approach never fills.
4. The site’s own key events, named as what they are
Form sends, call clicks, bookings, downloads: these are GA4 key events and we label them as GA4 key events. They are never called enquiries or leads in our reports, because an enquiry is something the client’s team qualifies, and the number they count is the number that matters. Where the client has its own terms for its pipeline, we use those terms and only those.
A note on Search Console, October 2026. Google is moving source links higher inside AI answers and adding an official-pages section at the top of some responses. None of that will show in your position data: Search Console reports every link in an AI Overview at the overview’s single position, and Google publishes no click data for AI answers. That is not a reason to ignore Search Console; it is the reason the four measures above exist.
What we will not do
- Report a single run of a prompt as a result, in either direction.
- Present a tool’s visibility score as a measure of anything other than that tool’s method.
- Claim that an AI mention caused a sale.
- Use a third-party authority score as if it were a Google or model metric.
- Promise a timescale. We state priorities and we report what moved.
How this fits the engagement
The measures map onto the stages of the Engagement Model. Momentum tells you whether you are being considered. Trusted-source mentions tell you whether you are becoming known. Human referrals tell you whether your content is worth sharing. Key events tell you whether any of it reached the site. None of the four replaces the client’s own count of enquiries and wins, which remains the only number that pays anyone.
To see where you stand before any of this is measured, start with the AI Visibility Audit.