§ 1 — Purpose and scope
What this measures
Generative search systems increasingly mediate how patients, physicians, executives and healthcare buyers identify organizations, providers, services and vendors. This protocol measures the extent to which a given organization participates in those answers, relative to the organizations it competes with.
1.1 In scope
The frequency with which a named organization appears in generative answers to a defined set of discovery questions, across defined platforms, within a defined geography and time window — and its position relative to a defined competitive set.
1.2 Out of scope
This is a visibility measurement. It is not a quality measurement, a volume measurement, or a causal one. Clause §9 states the limitations in full, and those limitations are reported alongside every figure produced under this protocol rather than appended as a disclaimer.
§ 2 — Definitions
Terms used only as defined here
The three appearance measures below are reported separately and are never combined into a single composite score.
- Discovery conversation
- One execution of one question on one platform at one recorded time. The atomic unit of measurement.
- Question set
- The fixed collection of questions defined for an engagement under §3, held constant across the measurement window and across subsequent re-measurements.
- Competitive set
- The organizations against which share is calculated. Defined before measurement begins from market knowledge, not derived from results — deriving it from results makes the measure circular. Organizations appearing in results but absent from the named set are recorded as the discovered set, and are frequently the most valuable finding in an engagement.
- Presence
- The organization is named anywhere in the answer body. Binary per conversation. Includes neutral and negative mentions; sentiment is recorded separately under §5.4.
- Recommendation
- The organization appears within the answer's recommended, shortlisted or affirmatively suggested set — not merely mentioned in passing, contrast or caveat. The only measure with direct acquisition value.
- Citation
- A domain owned by the organization appears among the answer's cited or linked sources. A leading indicator: citation share tends to precede recommendation share.
- Presence Rate
- The percentage of discovery conversations in which the organization is present. Absolute and independent of competitors. Ranges 0–100% and does not sum across organizations.
- Share of Discovery
- The organization's appearances as a proportion of all competitive-set appearances. Relative, and sums to 100% across the competitive set. Distinct from Presence Rate and never used interchangeably with it.
- Volatility
- The run-to-run instability of a question's results. High volatility indicates a contested, movable answer; low volatility indicates an entrenched one. See §6.3.
§ 3 — Question set construction
Built per engagement, then frozen
A question set that changes between measurement windows produces trend data that means nothing.
3.1 Categories
| Category | Measures | Minimum |
|---|---|---|
| unbranded | Core competitive discovery. Service or procedure plus geography, no organization named. | 8 |
| condition | Condition-led and procedure-led search paths. | 6 |
| comparative | Direct comparison between named organizations. | 4 |
| branded | Organization named. Scored for accuracy under §5.4, not for share. | 4 |
| physician | Individual provider discovery. Scored at physician level per §5.5. | 4 |
| quality | Outcomes, safety, ratings and accreditation framings. | 3 |
| access | Insurance acceptance, wait times, location, scheduling. | 3 |
| b2b-category | Vendor and solution category discovery. B2B engagements only. | 8 |
| b2b-evaluation | Selection-criteria and evaluation framings. B2B engagements only. | 4 |
Minimums per service line, per market. Provider engagements use the first seven categories; B2B engagements use comparative, branded and the two b2b categories.
3.2 Construction rules
- Questions are phrased as a person would actually ask them, not as keyword strings.
- Every unbranded question carries an explicit geographic qualifier. Geography defines the competitive set and cannot be left implicit.
- No question names the client organization except in the branded and comparative categories.
- Questions are recorded verbatim and republished in the report appendix, so any party can re-run them.
- Once a measurement window opens, the set is frozen. Additions begin a new baseline; they do not extend the existing one.
§ 4 — Sampling
Single-run measurement is not measurement
Generative systems are non-deterministic. Identical questions return different answers across runs, sessions, accounts, locations and model versions. Any figure derived from a single run is indefensible on challenge.
4.1 Run requirements
- Minimum five runs per question, per platform. The actual run count is stated in the report.
- Each run executes in a fresh session, unauthenticated, with memory and personalization disabled.
- Runs are distributed across at least three separate calendar days within the window, not executed consecutively.
- Geography is controlled to the target market by the same mechanism for every run, and the mechanism is recorded.
- Every run is stamped with platform, model version where exposed, timestamp and geographic setting. A run without a complete stamp is discarded.
4.2 Platforms
Four platforms constitute a complete measurement: ChatGPT, Google AI Mode and AI Overviews, Gemini, and Perplexity. A measurement covering fewer is reported as partial and names the omissions.
4.3 Weighting
Platform results may be weighted by usage share. Where weighting is applied, the report states the weights, their source, and the source's date. Where no defensible usage data exists, equal weighting is used and declared. Both weighted and unweighted figures are reported so a reader can judge the effect of the choice.
§ 5 — Scoring
Rules, not judgement
Each discovery conversation is scored independently against the three measures. Edge cases resolve by the rules below, so that two analysts scoring the same answer reach the same result.
| Situation | Presence | Recommended | Note |
|---|---|---|---|
| Named in the recommended set | Yes | Yes | — |
| Named as contrast or caveat | Yes | No | — |
| Named only inside a quoted directory listing | Yes | No | Source-mediated flag |
| Named with a factual error | Yes | Per placement | Accuracy flag, §5.4 |
| Named negatively | Yes | No | Sentiment flag, §5.4 |
| Affiliated physician named, organization not | No | No | Physician level, §5.5 |
| Facility named, system not named | Yes | Per placement | Roll up; record facility |
| Answer declines to recommend anyone | No | No | Counts in denominator |
Refusals and non-answers remain in the denominator. Excluding them inflates every rate.
5.2 Citation scoring
Citation is scored only where a domain owned by the organization appears among the answer's sources. Third-party pages describing the organization are not citations of it; they are recorded in the source landscape under §7.
5.3 Double-blind requirement for published work
Any measurement published as research is scored by two analysts independently, with disagreements resolved by documented adjudication. Client engagements may use single scoring, disclosed as such.
5.4 Accuracy and sentiment flags
Every presence event is additionally flagged for factual accuracy and sentiment. These do not affect share calculation. They aggregate into the reputation findings, which in health system engagements frequently carry more executive urgency than the share figures themselves.
5.5 Physician-level scoring
Individual providers are scored on presence and recommendation using the same rules, against a competitive set of individually named providers. Physician-level and organization-level results are reported separately and never aggregated together.
§ 6 — Calculation
How the figures are derived
6.1 Rates
For organization o, question q, platform p, over R runs: Presence Rate = runs where o present ÷ R Recommendation Rate = runs where o recommended ÷ R Citation Rate = runs citing o's domain ÷ R
6.2 Share of Discovery
Share of Discovery(o) = presence events for o
÷ presence events for all of the competitive set
Sums to 100% across the set.
6.3 Volatility
Volatility(q) = distinct organization sets returned ÷ R 1.0 = every run returned a different set (fully contested) 0.2 at R=5 = every run returned the same set (entrenched)
Volatility is an analytical output, not a diagnostic byproduct. High-volatility questions are the most movable — where the model has no settled answer, comparatively small changes in the source landscape can shift results. Low-volatility questions with a competitor entrenched are expensive to contest regardless of their commercial appeal.
6.4 Aggregation
Service-line and organization-level figures are the weighted mean of constituent question-level rates. Results are never averaged across geographic markets. Each market is reported separately; a system operating in four markets has four Share of Discovery pictures, and averaging them destroys the finding.
6.5 Reported precision
All rates are reported as observed ranges across runs alongside the central value, never as bare point estimates. Values are rounded to whole percentages. Where a rate derives from fewer than fifteen runs, it is marked provisional.
§ 7 — Source landscape
What explains the position
Share figures describe the position. The source landscape explains it, and it is the half of the report an executive can assign to an owner. Healthcare is distinctive here: it has a directory substrate no other sector has, and that substrate is where most citations originate.
7.1 Source capture
Every cited or linked source in every scored run is recorded by domain and classified: owned, provider directory, payer directory, government or regulatory, ratings and accreditation, clinical reference, news and editorial, review platform, social, or other.
The healthcare reference set includes at minimum NPPES and NPI records, Google Business Profile, Healthgrades, Vitals, Zocdoc, WebMD Care, US News, Leapfrog, CMS Care Compare, payer provider directories, health system service-line pages, and local news.
7.2 Discovered organizations
Organizations appearing in results but absent from the named competitive set are recorded, counted and reported. An organization crossing 10% presence across the question set is flagged for promotion into the competitive set at the next measurement, with the change logged. Promotion changes the share denominator and breaks strict comparability — the report states this wherever it occurs.
7.3 Entity consistency check
For each organization measured, its representation is compared across the directory substrate for consistency of name, address, specialty taxonomy, affiliation and service listings. Inconsistency is recorded as an entity clarity finding. This is typically the most concrete and most actionable output of an engagement.
§ 8 — Reporting standards
What every compliant report states
A report is compliant with this protocol only if it states all of the following. A report omitting any of them is not a Share of Discovery measurement.
- The protocol version used.
- The complete question set, verbatim, in an appendix.
- The named competitive set and the basis on which it was defined.
- Platforms covered, and any omitted.
- Run count per question per platform.
- Measurement window start and end dates.
- Model versions observed, where the platform exposes them.
- Geographic control mechanism.
- Weighting scheme and its source, or a declaration of equal weighting.
- Observed ranges alongside every central value.
- Single or double scoring per §5.3.
- The limitations in §9, in full, in the body of the report.
8.1 Comparability
Figures are comparable across measurement windows only where protocol version, question set, competitive set and platform coverage are unchanged. Any change is disclosed at the point of comparison, and comparisons across a change are labelled indicative rather than measured.
§ 9 — Limitations
Stated in every report, in full
Stating limitations openly is what distinguishes a measurement from a sales instrument.
§ 10 — Versioning and governance
How this document changes
10.1 Version policy
Versioned MAJOR.MINOR. A MAJOR increment changes how a figure is calculated and breaks comparability with prior results. A MINOR increment clarifies, extends categories, or adds reference sources without altering calculation. Every published figure carries its protocol version.
10.2 Change log
All changes are logged publicly with date, rationale and comparability impact. Silent revision of a published standard destroys the standard.
10.3 Open specification
This protocol is published in full. Third parties may apply it and are asked to cite the version used. Independent replication is treated as validation of the standard rather than as competition — a measurement standard that only its author can run is not a standard.
Measured against this protocol
A Share of Discovery assessment applies this specification to your organization — showing where you appear, where competitors are recommended instead, and which sources are producing those answers.