How search engines and AI assistants discover, retrieve, cite, and represent Medicare plan information.
People increasingly use search engines and AI assistants to answer questions about Medicare coverage. These systems can summarize plan choices, identify relevant types of coverage, explain benefits and costs, and direct consumers to resources for additional information.
But an important question remains largely unanswered:
Where do these systems get the Medicare facts used to construct their answers?
From Observation to Measurement
The Medicare AI Visibility Study is the measurement phase of a research program that began with observations of changing Medicare discovery patterns across search engines and AI systems.
In February 2026, the Trust Publishing Institute published The Medicare Digital Visibility & Compliance Shift: 2026–2028, a baseline research report examining how AI-mediated retrieval was changing Medicare plan discovery, interpretation, and publisher visibility.
The report was subsequently submitted as supplemental observational research to the federal record through public Requests for Information.
Among its central propositions were that conventional search visibility was becoming an incomplete measure of Medicare digital visibility, that Plan-ID entities provided a useful surface for observing AI retrieval behavior, and that Medicare organizations would increasingly need to monitor how search and AI systems retrieve and represent their information.
The report also proposed an industry-level Medicare Plan-ID visibility measurement that could compare plan and publisher visibility across conventional search and AI-driven retrieval systems.
The Medicare AI Visibility Study puts those propositions under measurement.
Rather than assuming that the observations and forecasts in the baseline report were correct, we built instrumentation to test them across a standardized population of Medicare entities.
The initial study follows 2,146 Medicare Plan-ID entities across major search and AI systems and measures how those systems discover, retrieve, rank, cite, and attribute information about the same underlying plans.
This creates a longitudinal record that can test a central question raised by the original research:
Is Medicare visibility actually shifting from page-based search visibility toward machine-mediated retrieval and answer visibility, and if so, who supplies the information those systems use?
Why We Are Studying Medicare AI Visibility
Medicare creates an unusually difficult information-retrieval problem.
Coverage varies by plan year, geography, plan type, eligibility, benefits, cost sharing, drug coverage, provider networks, and other factors. A fact can be accurate for one Medicare plan and completely wrong for another.
It is therefore possible for an AI-generated answer to contain individually accurate Medicare facts while assembling them in the wrong context.
During our 2026 research, we began observing another problem.
The publishers appearing prominently in conventional search results were not necessarily the publishers being retrieved or cited by AI systems. We also observed substantial differences in source selection between AI platforms.
This suggested that two measurements traditionally treated as related may need to be evaluated independently:
- Search visibility: whether a resource appears prominently in conventional search results.
- Answer visibility: whether a resource is retrieved, cited, or used as evidence within an AI-generated answer.
We created the Medicare Visibility Monitor to measure that distinction at scale.
The Study Population
The initial study population consists of 2,146 Medicare Plan-ID entities identified from Centers for Medicare & Medicaid Services data.
A Medicare Plan ID provides an unusually useful experimental unit because it identifies a specific Medicare plan rather than a broad topic or marketing keyword.
Instead of asking different systems unrelated Medicare questions and attempting to compare the results, the study repeatedly observes how multiple systems respond to queries about the same underlying Medicare entities.
This creates a standardized population for comparing publisher visibility across search engines and AI answer systems.
What We Measure
The Medicare Visibility Monitor records observable characteristics of search and AI responses for the study population.
Measurements include:
- whether a Medicare entity is discovered;
- which publisher resources appear in conventional search results;
- the search position of discovered resources;
- which publisher resources are retrieved by AI systems;
- which publisher resources are cited;
- citation frequency;
- differences in source selection among platforms; and
- changes in these measurements over time.
Later phases of the research examine the factual assertions contained within generated answers and the sources associated with those assertions.
Systems Under Observation
The study currently observes Medicare information across major search and answer environments, including:
- Google Search;
- Google AI-generated search experiences;
- Bing;
- ChatGPT; and
- Microsoft Copilot.
Platforms may be added, removed, or measured differently as their products and source-attribution systems evolve.
Research Questions
The Medicare AI Visibility Study is designed to investigate questions including:
- Does conventional search visibility predict AI citation visibility?
- Do different AI systems rely on the same publishers for information about the same Medicare plan?
- Which publishers are most frequently discovered, retrieved, and cited?
- How stable is publisher selection over time?
- Can a publisher have substantial answer visibility without substantial conventional search visibility?
- Can a publisher dominate conventional search results while contributing relatively little evidence to generated answers?
- Which sources are associated with specific factual assertions about Medicare plans?
- When publishers provide conflicting information, which assertions are selected by answer systems?
- How accurately do generated answers preserve plan year, geography, eligibility, plan type, benefits, and cost-sharing context?
- How does visibility change when the information available about a Medicare entity is materially improved?
An Early Observation: Search Visibility Is Not Answer Visibility
One of the earliest observations from the study is that conventional search visibility and AI answer visibility can differ substantially.
A publisher that rarely appears prominently in conventional search results may nevertheless be retrieved or cited frequently by an AI system. Conversely, a publisher with substantial conventional search visibility may have considerably less visibility within generated answers.
The pattern also varies by platform. A resource frequently selected by one answer engine may receive little or no visibility from another when both systems are asked about the same Medicare entity.
For this reason, the study does not treat search rankings, retrieval, citations, and factual attribution as interchangeable measurements.
From Citation Measurement to Assertion Measurement
Knowing which publisher receives a citation answers only part of the question.
The next phase of the Medicare AI Visibility Study examines the factual assertions contained within generated Medicare answers.
For an individual Medicare plan, those assertions can include identity, plan year, carrier, plan type, service area, eligibility, premiums, deductibles, maximum out-of-pocket limits, prescription drug coverage, cost sharing, benefits, and other plan characteristics.
This allows the research to move beyond asking:
Who received the citation?
and begin asking:
Who supplied the assertion?
That distinction is central to understanding how factual information moves from publishers into AI-generated answers.
Why Medicare Is a Useful Research Environment
Medicare provides characteristics that make this type of research unusually practical.
Thousands of plans have persistent identifiers. Multiple publishers describe the same underlying entities. Many important plan facts originate from authoritative government datasets. Plans have defined geographic and temporal boundaries. And many factual assertions can be compared against source data.
That gives researchers something that is difficult to obtain in many other information environments: a large population of identifiable real-world entities described by competing publishers and interpreted by multiple search and AI systems.
The objective is not simply to determine which website appears most often.
The larger objective is to understand how machines select among competing representations of the same underlying facts.
Study Methodology
The Medicare AI Visibility Study is an observational research project conducted on live search and AI systems.
The same population of Medicare entities is measured repeatedly so that changes in discovery, ranking, retrieval, citation, attribution, and publisher selection can be observed over time.
The project distinguishes observations from causal conclusions.
A change in visibility following a publishing change, search engine update, model update, or other event does not by itself establish that the event caused the observed result.
Where causal relationships cannot be established, findings are reported as observations rather than explanations of undisclosed ranking, retrieval, or model behavior.
Study Limitations
Search engines and AI systems are dynamic environments. Results may change because of model updates, index changes, retrieval systems, geographic context, personalization, interface changes, source-selection changes, or other factors unavailable to outside researchers.
Generated responses may also vary between repeated requests.
A citation does not establish that a source was used for model training, nor does the absence of a citation establish that a source had no influence on a generated response.
The Medicare Visibility Monitor measures observable system behavior. It does not provide access to the internal ranking, retrieval, training, or generation processes of the systems being studied.
These limitations are treated as part of the research rather than eliminated through assumptions about how a platform operates internally.
Research Timeline
September 2026 — Baseline Measurement
Initial measurement establishes cross-platform visibility for the Medicare Plan-ID study population and identifies differences among conventional search results and AI-generated answers.
2026–2027 — Longitudinal Observation
Repeated measurements will track changes in publisher discovery, search visibility, retrieval, citation, factual attribution, and source selection across the study population.
Medicarians 2027 — Initial Findings
Initial findings from the Medicare AI Visibility Study will be presented at Medicarians 2027.
The presentation will examine what the study reveals about search visibility, answer visibility, publisher selection, factual attribution, and the movement of Medicare facts through AI-generated answers.
Ongoing Research
Selected findings, methodology updates, and subsequent phases of the research will be published as the study continues.
Why This Matters
For consumers, the question is straightforward: Can I trust the Medicare information in the answer I am being shown?
For publishers, insurers, regulators, and technology companies, there is another question:
Which sources are shaping that answer?
As information discovery moves beyond lists of webpages toward generated answers, measuring where a website ranks is no longer sufficient to describe its influence.
We need to understand which resources machines discover, which resources they retrieve, which resources they cite, which factual assertions they select, and how accurately those assertions preserve the context of the underlying entity.
The Medicare AI Visibility Study was created to measure that emerging information environment.
About the research: The Medicare AI Visibility Study is an ongoing observational research project conducted by MedicarePlans.com. Study methods and measurements may evolve as search engines and AI platforms change their products, interfaces, and source-attribution systems. For more information, contact David Bynon at david@medicarewire.com