What metrics measure AI search visibility for a hospital?

Five. ① Mention rate: share of a fixed question set on which the hospital is named (branded questions excluded). ② Per-engine breakdown: the same set on each of four engines. ③ Share of voice: the hospital's share of all hospital names mentioned (names merged first). ④ Rank-weighted score: position-weighted, secondary. ⑤ Citation: whether the hospital's own document is a source, and the source types. There is no ranking, so these five stand in for ‘rank’. Boily uses them in its biweekly reports.

Can I know my hospital's position in an AI's recommendation ranking?

Not as a rank, but as metrics. Answers are prose and vary run to run, so ‘No. 3’ does not exist. Instead: mention rate for how often you appear, share of voice for your share among local hospitals, rank-weighted score for whether you are near the top when mentioned. Rounds must accumulate before a direction is visible.

How is share of voice calculated?

Count every hospital name mentioned across the round's answers and take your share. An answer naming three hospitals contributes 3 to the denominator. Normalise names first — full, short, branch and district-prefixed variants of the same hospital must be merged, or every hospital looks smaller than it is (193 raw → 70 in one dental case). Exclude branded questions and report per engine.

How do I verify the numbers in a vendor's report?

Ask four things: were questions containing the hospital's name in the denominator (they inflate); were four engines blended into one (hides a single-engine surge); were competitor names normalised (otherwise share of voice is wrong); and is the question set identical every round. If all four are right the numbers are usable; if any is not, round-to-round comparison does not hold.

Metrics · Five numbers

AI search visibility metrics for hospitals — the five numbers that replace “rank”

Published 2026-09-26 · Boily, Korea

AI search has no ranking list, so a hospital's position is measured with five numbers: mention rate, per-engine breakdown, share of voice, rank-weighted score and citation. Each answers a different question, and blending them is the most common way to misread a report. Boily uses these five in its biweekly reports and guarantees no visibility or ranking.

The five metrics — what is counted, how, and how to read it

1. Mention rate

What —: The share of a fixed question set on which the hospital's name appears in the answer. Questions containing the hospital's own name are excluded from the denominator (they almost always hit and inflate the number).
Calculation —: Mentioned (question × engine) cells ÷ all cells. E.g. 40 questions × 4 engines = 160 cells, 32 mentioned → 20.0%.
Reading —: The first number to look at and the only one that supports round-to-round comparison. A single round's move is noise; judge on three or more.
Example —: Boily's own round 8: 43/384 = 11.2%.

2. Per-engine breakdown

What —: The same set seen separately on ChatGPT, Claude, Gemini and Perplexity. Four numbers, never blended.
Calculation —: Per engine: mentioned questions ÷ questions.
Reading —: Equal totals with different composition are different states. Reading one engine's surge as overall improvement is the most common misreading; ‘ahead’ is claimed only when a majority of engines moved.
Example —: Boily's own round 8: Perplexity 39.6% · Gemini 4.2% · ChatGPT 1.0% · Claude 0.0%. The 11.2% total is almost one engine's number.

3. Share of voice

What —: Of all hospital names mentioned across a round's answers, the hospital's share. ‘Who occupies the answers’.
Calculation —: Own mentions ÷ all hospital mentions. An answer naming three hospitals adds 3 to the denominator. Full names, short names and branch names are merged first — otherwise one hospital fragments into several.
Reading —: Explains why mention rate is low: a few hospitals hold the slots, nobody does, or the engine names no hospitals for that question. Three situations, three different responses.
Example —: In one dental measurement 193 raw names normalised to 70. Without that step share of voice is simply wrong.

4. Rank-weighted score

What —: When mentioned, how far up the answer the hospital appeared. A first mention and a passing line count differently.
Calculation —: Per engine and question: not mentioned 0 · mentioned 0.3 · upper part 0.7 · first 1.0; summed across engines as a percentage of the maximum. Always ≤ mention rate.
Reading —: A secondary metric for hospitals whose mention rate has plateaued: is the position moving up? Position fluctuates between rounds, so never judge on this alone. Hospitals whose baseline is mention rate keep comparing on mention rate.
Example —: One dental clinic, same round: mention rate 52.6% · rank-weighted 35.4. Mentioned, but not near the top.

5. Citation and source type

What —: Whether the hospital's own document appears as a source of the answer, and what kind of sources appear (hospital site, map, directory, competitor site).
Calculation —: Share of mentioned answers whose sources include the hospital's domain + a tally by source type.
Reading —: Separates ‘named, but the link goes elsewhere’. If visibility exists but visits do not, this is usually why. Which document types get cited decides what to build next.
Example —: Boily's own vendor set: Perplexity mentioned Boily on 7 of 12 questions yet cited boily.co.kr on 0 calls. Mention and citation are different metrics.

Which metric, when

First measurement —: Mention rate and per-engine split. Set the baseline and see which engine is at zero.
Re-measurement after remediation —: Direction of mention rate (3+ rounds) plus per-engine split: one engine or a majority?
When mention rate plateaus —: Share of voice and rank-weighted score: are the slots empty or taken, is the position rising?
Visibility without patient visits —: Citation: named, but linked elsewhere?
Reading a vendor's report —: Ask which of the five was calculated and how: branded questions in the denominator? engines blended? names normalised?

FAQ

Q. What metrics measure AI search visibility for a hospital?: A. Five. ① Mention rate: share of a fixed question set on which the hospital is named (branded questions excluded). ② Per-engine breakdown: the same set on each of four engines. ③ Share of voice: the hospital's share of all hospital names mentioned (names merged first). ④ Rank-weighted score: position-weighted, secondary. ⑤ Citation: whether the hospital's own document is a source, and the source types. There is no ranking, so these five stand in for ‘rank’. Boily uses them in its biweekly reports.
Q. Can I know my hospital's position in an AI's recommendation ranking?: A. Not as a rank, but as metrics. Answers are prose and vary run to run, so ‘No. 3’ does not exist. Instead: mention rate for how often you appear, share of voice for your share among local hospitals, rank-weighted score for whether you are near the top when mentioned. Rounds must accumulate before a direction is visible.
Q. How is share of voice calculated?: A. Count every hospital name mentioned across the round's answers and take your share. An answer naming three hospitals contributes 3 to the denominator. Normalise names first — full, short, branch and district-prefixed variants of the same hospital must be merged, or every hospital looks smaller than it is (193 raw → 70 in one dental case). Exclude branded questions and report per engine.
Q. How do I verify the numbers in a vendor's report?: A. Ask four things: were questions containing the hospital's name in the denominator (they inflate); were four engines blended into one (hides a single-engine surge); were competitor names normalised (otherwise share of voice is wrong); and is the question set identical every round. If all four are right the numbers are usable; if any is not, round-to-round comparison does not hold.

Your hospital's five numbers come from the first measurement

Boily measures Korean hospitals biweekly on a fixed, hospital-specific question set across four engines and reports all five. No visibility or ranking is guaranteed.

Visibility benchmark Case records Advertising law review

Written by Boily. Definitions are those used in Boily's own measurement and may differ from other vendors' usage. Examples are Boily's measurements; hospital names are withheld. Boily guarantees no visibility or ranking.