How the Medicare AIVI™ Index Is Built, Scored, and Corrected

1. Document version and status

This is Medicare AIVI™ Methodology v1.0, effective 2026-10-01. It is the controlling document for the Q3 2026 Medicare AI Visibility Benchmark and for every derived analysis published from it, including the UnitedHealthcare analysis and the Kaiser Permanente analysis.

Two rules govern this document going forward. First, a change to the QVI™ formula, to the attribution rules, or to the surface set is a major version bump, and indices scored under different major versions are not directly comparable. Second, corrections to this document are additive: superseded text is recorded in the version history above rather than deleted.

2. What Medicare AIVI™ measures

Medicare AIVI™ answers one question: when a Medicare beneficiary asks an AI assistant a question in this category, how present, how early, and how favorably does this plan or carrier appear in the answer? It is a measure of the answer layer specifically. It is not a search ranking, not a Star Rating, not an enrollment estimate, and not a brand-health survey.

The index deliberately reports two measures separately rather than blending them into one score:

• Share of Voice (reach). How much of the conversation belongs to a brand, reported against two denominators: Commercial share (against the 59,055 scored commercial rows) and Total entity share (against the 229,372 total extraction rows). Neither is a direct estimate of demand, attention, traffic, enrollment, or revenue.
• QVI™ (quality). How strongly each individual appearance performs, independent of volume. This is the deliberate departure from the Insurance AIVI™, which blends Share of Model into its composite at 35% weight. A Medicare QVI™ figure is therefore not directly comparable to an Insurance AIVI™ figure.

Four things are deliberately outside the index:

• Paid placements. Sponsored and ad-click results are excluded from the score entirely. A plan cannot buy QVI™ points.
• Traffic and enrollment. The index measures presence in the answer, not what a reader does next or whether they enroll.
• Star Ratings and plan benefit design. QVI™ scores how an AI answer treats a brand, not CMS Star Ratings, premiums, or benefit structure.
• Private or client-supplied data. Every figure in a published index is derived from public model outputs. No client relationship contributes to any score.

3. Study universe and prompt set

An index is only as defensible as the question set that produced it, so the construction is fixed before any responses are collected and locked before scoring.

Two denominators, deliberately kept apart. Use 14,490 for overall visibility/availability measures — a non-render counts as zero brand visibility on that query, for every brand. Use 14,405 for analysis of what is inside generated content: entity extraction, the Total Visibility Universe, Share of Voice, and everything built on it. Both are reported so neither figure is read as the only correct one.

Google AI Overview non-renders are a real outcome, not missing data. AI Overview rendered for 95.89% of its allocated prompts (1,985 of 2,070); the other 85 are observed non-render outcomes, carried as zero-visibility in the 14,490-denominator figures.

Response coverage. Computed against the canonical 115-brand list from sealed per-surface data: 11,638 of 14,405 populated responses (80.79%) name at least one scored commercial brand; against the full 14,490 designed sample with non-renders folded in as zero-visibility, the figure is 80.32%.

Entity classification and rollups. Name variants are rolled up under one canonical brand — for example, state Blue Cross Blue Shield affiliates (Illinois, Texas, Georgia, and unbranded “BCBS” mentions) roll up to Blue Cross Blue Shield Medicare. Brand references are separated from generic Medicare vocabulary, and consolidated tool and phone references (Medicare Plan Finder, 1-800-MEDICARE) are each treated as a single entity rather than counted as separate variants.

Reliability threshold. QVI™, mean recommendation score, own-domain citation rate, top-3 rate, and mean position score are published only for brands with at least 100 mentions — a reporting rule, not a statistical confidence guarantee. Mentions and Commercial share are complete for every brand regardless of this threshold; small-volume QVI™ scores below it can be extreme and should not be read as robust leaders.

4. Attribution rules

Row-based, not response-based

59,055 scored rows underlie every commercial-share and position figure in this report. QVI™ and position scores remain row-based; repeated mentions of one brand within a response can occupy several positions, and a brand's mention count is a count of scored rows, not of distinct responses it appeared in.

Scored-subset marking (†)

QVI™, mean recommendation score, own-domain citation rate, top-3 rate, and mean position score are computed on the subset of rows that received full per-mention scoring. For brands marked † on the leaderboard, these five columns reflect that scored subset rather than the brand's complete mention count; mentions and Commercial share remain complete for every brand regardless.

Recommendation coverage

All 23,715 rows from the four API configurations (ChatGPT, Gemini, Claude, Perplexity) have Recommendation Score zero. Nonzero scores occur only on consumer-facing surfaces (Google AI Overviews, Microsoft Copilot, Bing Copilot). QVI™ can therefore reflect surface mix and incomplete recommendation signals as much as brand performance.

5. The QVI™ formula

QVI™ formula. Quality Visibility Index = 8 × mean Position Score + 7 × mean Recommendation Score + 25 × own-domain citation rate. Citation rate enters as a fraction from 0 to 1. The theoretical maximum is 100. QVI™ is a per-mention average, not a volume count. It is deliberately volume-neutral, unlike the Insurance AIVI™ score, so a brand's rate of being answered well can be read separately from how often it is answered at all (Commercial share, Section 3).

6. Known limitations

These are the constraints a reader should hold in mind when using a Medicare AIVI™ figure. They are published here rather than buried in a footnote because a benchmark that only lists its strengths is marketing.

Capture quality

Some Microsoft Copilot records contain only prompt/location text, an interface greeting, or an unfinished response indicator. Populated is therefore a storage test, not a substantive-answer test.

Recommendation Score is not uniform across surfaces

All 23,715 API-surface rows have Recommendation Score zero; this component is not comparable as a uniformly measured cross-surface preference signal. A brand's QVI™ can be shaped by which surfaces it is captured on as much as by how it is treated within them.

Model responses are non-deterministic

Across 7,363 observed brand/query/surface combinations, 52.0% are not present in all three reported runs, and 40.6% show a rank swing of at least two positions when consistently present. Single-answer visibility claims are not a stable baseline; repeated observations support a more useful descriptive one.

Scope

This benchmark does not validate extraction accuracy beyond the classification rules stated here, resolve every plan alias, or rescore recommendations. Prominence, Product Ownership, Geography, Query Ownership, Co-Mention Network, and Cross-Surface Agreement analyses require response-level, product-, geography-, or query-tagged brand data beyond this benchmark's current scope and are not included.

An edition is a snapshot

Data were captured 2026-06-29 to 2026-07-03, crossing the Q2/Q3 boundary; the workbook retains the Q3 study label as a convention, not a claim that every observation fell inside Q3. Surfaces update their retrieval stacks and system prompts continuously and without notice, so an edition describes the answer layer during its collection window and nothing outside it.

Category scope

Scores are normalized within the Medicare study cohort. A QVI™ figure here and an Insurance AIVI™ figure must not be compared — the two indices weight volume differently by design (Section 2).

Independence

Brainpan.AI publishes AIVI™ indices independently. Being named in an index — favorably or otherwise — reflects observed public model output and nothing else. Brands cannot pay for inclusion, placement, or removal, and no client relationship has ever influenced a published figure.

7. Correction policy

Any organization named in a published Medicare AIVI™ index can ask for a figure to be corrected. This is a standing commitment, not a courtesy, and it applies whether or not the organization is a Brainpan.AI client.

What qualifies

• Factual error — a miscount, a transcription error, a brand misattributed to the wrong corporate entity, or a figure that does not reconcile with the published data.
• Methodological misapplication — a rule in this document applied incorrectly to the organization's data, for example a paid placement counted as organic.
• Entity error — the wrong legal entity, subsidiary, or brand family associated with a set of mentions.

What does not qualify

Disagreement with the methodology itself is not a correction request; it is a methodology comment, and it is welcome as such, but it is answered in a version of this document rather than by changing a score. A request to remove an unflattering but accurately measured figure will be declined and the exchange noted.

How to file

Send the request to kwalsh@brainpan.ai with the subject line AIVI correction request, naming the index edition, the specific figure, and the basis for the challenge.

8. References and related documents

• Q3 2026 Medicare AI Visibility Benchmark — the index this version governs.
• Medicare AIVI™ Benchmark — the series hub.
• AIVI™ Insurance Benchmark — the Insurance edition of this same measurement discipline.
• Multi-Engine AIVI™ Benchmarks — every industry the series covers, compared side by side.
• UnitedHealthcare Medicare analysis — a worked application of the scoring model to the commercial-share leader.
• Kaiser Permanente analysis — a second worked application, isolating the citation-authority term of the score.
• Glossary — canonical definitions for every metric named here.
• Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024) — the foundational academic treatment this prompt design follows.
• Schema.org Dataset — the vocabulary used to declare each published index as a machine-readable dataset.

Get the full index

The Q3 2026 Medicare AI Visibility Index scores all 115 visible brands on the components described above, with per-surface detail and the non-commercial layer. $3,500, one-time.

Buy the full index →

Single-organization commercial license / One-time payment — delivered within one business day.

Frequently Asked Questions