How Share of Answer is measured.
One number, one formula, one rule about sample size. If a report cannot show its working, it is not a measurement — so this page is the working, published.
The formula
shareOfAnswer = round(100 * named / total) named = runs where the client was mentioned total = runs that returned a usable answer total >= 3 or no percentage is shown
A mention is counted when the extractor finds the client's brand or domain in the answer text. Position is recorded when the answer ranks the business, and sentiment is classified from the sentence around the mention. All three are stored per run, so any number in a report can be traced back to the exact answer that produced it.
Why three runs is the floor
AI answers vary by model, day and phrasing. One run is an anecdote; two is a coincidence. Below three usable answers AudiFolio refuses to print a percentage and says so on the page. Between three and nine it prints the number with a caution. At ten or more it prints the date range and the sample.
| Runs | What the report says |
|---|---|
| 0–2 | “Not enough answers yet — AudiFolio needs 3 runs before it reports a share.” |
| 3–9 | “Based on N answers. Treat as a direction, not a measurement.” |
| 10+ | “Based on N answers collected between DATE and DATE.” |
Branded and non-branded questions are judged separately
A question that contains the client's name — “is ApexScale good for enterprise cloud” — cannot demonstrate discovery. It is tracked, and it is reported, but when there are at least three non-branded answers the headline comes from those. The branded score appears as a note beside it.
The panel
Answers come from an open-model panel behind one router, and every run records the model that produced it. When the plan allows a larger model, the report names it. AudiFolio does not silently swap engines mid-period: a change of model in the panel starts a new comparison baseline, and the report says which model answered which question.
The shared answer cache
The same question, asked in the same city, in the same language, is asked once and analysed separately for every client that tracks it. That is why the price can stay low: cache reuse can reduce repeated provider work. A cache hit still produces a fresh per-client run record, so each agency's history is its own.
What AudiFolio does not claim
- It does not promise a rank, a placement, or a citation.
- It does not claim coverage of a named assistant or search surface.
- It does not merge different surfaces into one number without saying so.
- It does not estimate a share when there is not enough evidence — it says so instead.
What you can verify
Every report page carries the questions, the answers, the timestamps, the model and the cited URLs. A client can read the evidence page and disagree with the interpretation, but not with the record. That is the point: the report is auditable, and the agency's argument rests on the record rather than on the tool.
What exactly does AudiFolio measure?
Share of Answer: the percentage of tracked AI answers that name your client. It is written as "named in 3 of 10 answers" beside the percentage, with the date range and the model that produced each answer. AudiFolio never promises a ranking.
How is this different from an AI visibility dashboard?
AudiFolio produces a four-page report with your logo, colour, dated evidence and briefs tied to the observed gaps. The client can review the shared report without a workspace account.
Can I put my agency's logo on it?
Yes. Upload a logo and set a colour once in Settings; every report and every share link uses them. The masthead reads "AudiFolio", with your agency credited as the publisher.
How much does it cost per client?
Plan pricing and included limits are listed on the pricing page. Your workspace displays the options currently available to your account.
Do you store my client's customer data?
No. AudiFolio stores the client domain, the questions you write, and the AI answers. There is no end-customer personal data anywhere in the system, and deleting a client deletes its runs and reports.
What happens when AI answers change?
They do change, which is why every question runs more than once and the report shows "observed in N of M" with dates. You are selling a trend with evidence, not a screenshot.