Healthcare Stocks and AI Search: Which Companies Are Gaining or Losing AI Recommendation Visibility?

Research status: Exploratory longitudinal research. AI recommendation momentum has not been validated as a predictor of healthcare revenue, membership growth, prescription volume, diagnostic-testing demand, analyst revisions, valuation, or stock returns.

Initial observation window: July through September 2026

Healthcare research slice: 5 mapped public parents

AI platform families: ChatGPT, Gemini, Google AI Mode, Google AI Overviews, Microsoft Copilot, and Perplexity

Current methodology version: V0

Answer Capsule

The initial LLM Authority Index healthcare research slice shows a mixed pattern in AI recommendation momentum from July to September 2026 rather than one uniform sector direction.

Among five mapped public-company parents, Molina Healthcare was the only positive aggregate mover, increasing recommendation coverage 3.7 percentage points, from 23.5% to 27.2%, but its exploratory interval crossed zero and its platform direction was mixed. Cigna was exactly flat on aggregate recommendation coverage at 35.2% in both periods. UnitedHealth Group / UnitedHealthcare declined 4.6 points, but its exploratory interval narrowly crossed zero. CVS Health, measured through CVS Pharmacy, declined 9.4 points and was the only company in this healthcare slice to meet the V0 Negative AI divergence candidate rules. Labcorp, measured through Labcorp OnDemand, declined 10.0 points, but the row remains Exploratory because it contains only 70 matched cells and five measured platform families.

Across the five-company slice, the median recommendation change was approximately -4.6 percentage points, and the simple mean was approximately -4.1 points.

These measurements describe changes in AI recommendation behavior only. They do not establish changes in medical membership, pharmacy utilization, prescriptions, diagnostic-testing volume, patient demand, premiums, revenue, earnings, or stock performance.

The prospective investor question is narrower: if changes in unbranded AI recommendation visibility persist, do they precede measurable changes in consumer consideration, branded search, site traffic, enrollment, pharmacy demand, diagnostic testing, analyst expectations, or reported financial outcomes?

That question sits inside the AI Commercial Momentum Hypothesis. The company-level measurement rules are documented in How We Measure AI Commercial Momentum, while the complete initial public-company panel is preserved in Initial Findings From 25 Public Companies.

Healthcare AI Recommendation Momentum at a Glance

This is a research-defined healthcare slice, not a formal healthcare equity index.

The companies represent different consumer journeys and economics. Molina and UnitedHealth Group are health-plan businesses. Cigna spans health-benefit and pharmacy-benefit exposure. CVS is measured through the CVS Pharmacy brand within a much broader healthcare parent. Labcorp is measured through Labcorp OnDemand, a consumer testing brand rather than the entire laboratory-services business.

Those differences are central to interpretation.

Molina, Cigna, and UnitedHealth Group also appear naturally in the Insurance Stocks and AI Search analysis. Their inclusion in two sector views reflects overlapping consumer decision journeys, not double-counting inside the company-level signal itself.

Questions This Section Answers

• Which healthcare companies gained or lost AI recommendation visibility?
• Which movements were broad across multiple AI platforms?
• Does a negative healthcare-sector median imply weaker healthcare fundamentals?

What the Company-Level Signals Show

Molina Healthcare: positive aggregate movement, but mixed platform evidence

Molina Healthcare increased recommendation coverage from 23.5% to 27.2%, a gain of 3.7 percentage points.

The exploratory 95% interval ranged from approximately -1.8 to +9.2 points, so it crossed zero. Platform direction was also mixed:

• ChatGPT: -12.50 pp
• Gemini: -7.14 pp
• Google AI Mode: +10.34 pp
• Google AI Overviews: +5.26 pp
• Microsoft Copilot: -4.35 pp
• Perplexity: 0.00 pp

Molina therefore remains Mixed / neutral under the V0 rules.

Presence increased 6.5 percentage points, from 40.6% to 47.0%, which was larger than the 3.7-point increase in recommendation coverage. That difference is important. Molina appeared in more eligible responses, but a larger presence increase does not automatically mean a proportionate improvement in recommendation status.

The correct conclusion is narrow: Molina's measured recommendation coverage increased modestly, but the evidence was not broad or directional enough to classify the row as a positive candidate.

Cigna: flat aggregate recommendation coverage, different platform stories underneath

Cigna recorded 35.2% recommendation coverage in both the base period and September, producing a 0.0 percentage-point aggregate change.

The platform-level changes were not all zero:

• ChatGPT: +9.38 pp
• Gemini: 0.00 pp
• Google AI Mode: +0.83 pp
• Google AI Overviews: 0.00 pp
• Microsoft Copilot: -4.08 pp
• Perplexity: -4.26 pp

This is a useful example of why an aggregate result can hide platform redistribution.

Cigna also improved average recommendation rank from approximately 4.38 to 4.03, while aggregate recommendation coverage stayed flat. Presence increased 1.5 points.

The public-parent row combines Cigna and Express Scripts under The Cigna Group. That mapping is economically reasonable for a parent-company research panel, but the measured consumer exposure is not identical to every segment of The Cigna Group.

Cigna therefore illustrates three separate measurements at once: flat recommendation coverage, slightly higher presence, and a numerically better average rank among observed recommendations.

These variables should not be collapsed into one AI visibility score.

UnitedHealth Group: a near-directional decline that remains inconclusive

UnitedHealth Group / UnitedHealthcare declined from 64.5% to 59.8% recommendation coverage, a change of -4.6 percentage points.

The exploratory interval ranged from approximately -9.5 to +0.3 points. The upper bound remained slightly above zero, so the current row is not classified as directionally negative.

Four platforms declined:

• ChatGPT: -11.11 pp
• Gemini: -21.88 pp
• Google AI Mode: +4.04 pp
• Google AI Overviews: -4.05 pp
• Microsoft Copilot: 0.00 pp
• Perplexity: -17.65 pp

The row is therefore close to, but does not satisfy, the V0 negative-candidate threshold.

The parent-company caveat is material. The measured entities are UnitedHealthcare, UnitedHealthcare Golden Rule, and UnitedHealthcare Vision. UnitedHealth Group is a much broader organization, including businesses that are not represented directly by those consumer-facing recommendation measurements.

The research should therefore describe this as UnitedHealthcare-related AI recommendation exposure mapped to UnitedHealth Group, not as a complete measurement of every economic driver inside UNH.

CVS Health: the only negative candidate in the healthcare slice

CVS Health declined from 34.7% to 25.3% recommendation coverage, a change of -9.4 percentage points.

The exploratory interval remained below zero, from approximately -17.2 to -1.6 points. Four platforms worsened, one improved, and one was stable:

• ChatGPT: +16.67 pp
• Gemini: -20.83 pp
• Google AI Mode: 0.00 pp
• Google AI Overviews: -8.00 pp
• Microsoft Copilot: -15.00 pp
• Perplexity: -43.75 pp

The row therefore satisfies the V0 negative-candidate rules.

Presence also declined, but by a smaller amount: -4.7 percentage points, from 78.2% to 73.5%. Average recommendation rank moved from approximately 2.72 to 3.46, which is numerically worse.

The economic mapping limitation is substantial. The V0 row measures CVS Pharmacy exposure and maps it to CVS Health Corporation. CVS Health also contains businesses and revenue streams that are not directly captured by pharmacy recommendation behavior.

The conclusion is therefore limited to the observed entity: CVS Pharmacy recommendation coverage declined materially across the matched AI prompt panel.

That does not establish a decline in CVS Health revenue, Aetna membership, Caremark activity, prescription volume, store traffic, or future stock performance.

Labcorp: a large decline with an Exploratory evidence base

Labcorp declined from 17.1% to 7.1% recommendation coverage, a change of -10.0 percentage points.

The magnitude is large, but the exploratory interval is wide, from approximately -22.0 to +2.0 points, and the row has only 70 matched cells across 56 prompt clusters.

Only five platform families contain measured values:

• ChatGPT: 0.00 pp
• Gemini: -25.00 pp
• Google AI Mode: +14.29 pp
• Google AI Overviews: -26.09 pp
• Microsoft Copilot: 0.00 pp
• Perplexity: unavailable

The V0 confidence classification is therefore Exploratory.

Labcorp is another important metric-separation example. Presence was exactly flat at 37.1%, while recommendation coverage fell 10.0 points. Average recommendation rank improved from 3.0 to 2.6 among the responses where Labcorp OnDemand was recommended.

That combination means Labcorp OnDemand was still being mentioned at roughly the same rate, and ranked somewhat better when recommended, but was recommended in a smaller share of eligible matched cells.

The parent mapping is also narrow. The measured entity is Labcorp OnDemand, not the entire Labcorp Holdings enterprise.

Why Healthcare Is a Plausible Sector for AI Commercial-Momentum Research

AI-mediated health discovery is already common enough to justify measurement, even though that does not establish a financial forecasting relationship.

KFF's June 2026 Tracking Poll on Health Information and Trust reported that 29% of U.S. adults said they use AI tools or chatbots for health information and advice at least monthly, up from 17% roughly two years earlier. KFF's March 2026 polling also found that 32% of adults had turned to AI chatbots for health information in the prior year. Common uses included looking up symptoms or conditions, understanding medical tests or diagnoses, comparing treatment options, and deciding whether to seek care.

Sources:

• KFF Tracking Poll on Health Information and Trust: Use of Social Media and AI For Health Information and Advice
• KFF Tracking Poll on Health Information and Trust: Use of AI For Health Information and Advice

This supports a narrow premise: AI systems participate in health information and decision journeys at meaningful scale.

It does not prove that recommendation visibility for Molina, Cigna, UnitedHealthcare, CVS Pharmacy, or Labcorp OnDemand predicts revenue or stock returns.

That remains a prospective empirical question.

Questions This Section Answers

• Why might platform breadth matter especially in healthcare?
• How does healthcare recommendation portability differ from citation-source portability?
• What should investors avoid assuming from one-platform healthcare movement?

Cross-Platform Portability Matters, but It Is a Separate Variable

Healthcare provides several examples of why aggregate recommendation change should not be interpreted without platform context.

CVS declined on four platforms, improved on ChatGPT, and was flat on Google AI Mode.

UnitedHealth Group / UnitedHealthcare declined on four, improved on one, and was flat on one.

Molina improved on two, declined on three, and was flat on one.

Labcorp OnDemand had only five measured platforms and split between negative, positive, and flat results.

These patterns fit the measurement principle developed in Does Cross-Platform AI Visibility Matter?: platform breadth should be measured separately from aggregate movement.

The existing LLM Authority Index Persistence-Portability Gap research provides useful context at the citation-source level. In the Health and Medical source studies, median monthly Top 100 source persistence was approximately 61.3%, while median cross-platform Top 100 portability was approximately 25.8%, producing an approximate 35.5 percentage-point gap for that source-market research.

See The Persistence-Portability Gap.

Those values are citation-source metrics, not company recommendation metrics. They should not be imported directly into this investor panel.

The conceptual lesson is narrower: healthcare AI ecosystems can be materially platform-specific, so one-engine movement should not automatically be treated as portable evidence across the AI discovery market.

Recommendation Coverage, Presence, and Rank Tell Different Stories

The healthcare slice contains several useful examples of metric divergence.

This is why AI Recommendations vs. Mentions vs. Citations treats recommendation frequency, presence, rank, citations, and sentiment as distinct measurements.

For investors, this distinction matters because downstream business mechanisms could differ.

A company that is mentioned just as often but recommended less frequently may be experiencing a different AI-discovery shift from a company that disappears from answers entirely.

A company that is recommended less often but ranks better when recommended presents yet another pattern.

The V0 research does not assume that any one pattern is financially superior. Those relationships have to be tested.

Why This Matters for Investors and Analysts

The practical value of this sector research is not a healthcare stock ranking.

It is the creation of a dated, company-level record that can later be matched against real operating outcomes.

Potential healthcare validation targets differ materially by business model.

The future validation should compare AI variables with these outcomes using only information available at the original signal date.

The prospective test design is described in Can AI Search Visibility Predict Revenue Growth?, and the broader backtesting architecture is documented in How Investors Could Backtest AI Search Signals.

What This Does Not Mean

The healthcare findings should not be interpreted as:

• evidence that CVS Health is financially deteriorating;
• evidence that Molina is gaining healthcare market share;
• evidence that UnitedHealth Group will miss revenue expectations;
• evidence that Labcorp testing demand is declining;
• evidence that Cigna is commercially unchanged;
• a healthcare stock ranking;
• a buy, sell, or hold recommendation;
• proof that AI systems are causing changes in patient or consumer behavior.

The current classification system identifies unusual AI recommendation momentum under a frozen research methodology.

The term Negative AI divergence candidate means only that the AI-side recommendation change met the V0 magnitude, interval, and platform-breadth thresholds. It does not imply that a stock is overvalued, that fundamentals are weakening, or that a negative return should follow.

Methodology

The healthcare sector article uses the same V0 company-level methodology as the broader AI Investor Signal Tracker.

Matched-panel denominator

For each public parent, the primary change compares the same normalized prompt on the same AI platform family in the base month and September 2026.

Recommendation coverage is:

recommended matched prompt-platform cells / eligible matched prompt-platform cells

Explicit company rows coded as not recommended remain in the denominator. Explicit extraction failures are excluded rather than treated as zero visibility.

Duplicate handling

Exact repeated response states exported into multiple vertical datasets are collapsed in the primary cleaned panel. A no-dedupe sensitivity result is retained for each public parent.

Parent-company mapping

Known brands and entity variants are rolled up to the mapped public parent for investor analysis.

Important healthcare mappings include:

• Cigna + Express Scripts -> The Cigna Group
• UnitedHealthcare + Golden Rule + UnitedHealthcare Vision -> UnitedHealth Group
• CVS Pharmacy -> CVS Health
• Labcorp OnDemand -> Labcorp Holdings

These mappings improve public-company analysis but do not imply that measured consumer exposure represents every business segment of the parent.

Exploratory interval

The primary point estimate is cell-weighted. For uncertainty, matched-cell changes are averaged within normalized prompts and a normal 1.96 interval is applied using the standard error of the prompt-level means.

This is an exploratory uncertainty interval, not a formal causal confidence interval.

V0 candidate rules

A Positive AI divergence candidate requires:

• recommendation change of at least +5 percentage points;
• interval lower bound above zero;
• at least four improving platform families.

A Negative AI divergence candidate requires:

• recommendation change of at least -5 percentage points;
• interval upper bound below zero;
• at least four worsening platform families.

Everything else is Mixed / neutral.

Confidence rules

High confidence requires at least 200 matched cells, six platform families, a directional interval, and small cleaning sensitivities under the frozen V0 thresholds.

Medium requires at least 100 matched cells and at least five platforms.

Rows below those thresholds are Exploratory.

That is why Labcorp remains Exploratory despite a -10.0-point recommendation change.

Limitations

1. The observation window is short, July through September 2026.
2. The healthcare slice contains only five mapped public parents and is not a representative healthcare equity index.
3. Prompt panels are research panels, not search-volume-weighted market samples.
4. Company recommendation coverage does not measure patient volume, prescriptions, membership, revenue, or market share.
5. Health-plan companies overlap with the insurance sector by design.
6. CVS Health is measured through CVS Pharmacy exposure, not every CVS Health business.
7. Labcorp is measured through Labcorp OnDemand and has only five platform families in the matched row.
8. UnitedHealth Group is measured through UnitedHealthcare-related consumer brands, not the full parent-company operating mix.
9. Cigna includes Cigna and Express Scripts entity exposure, which does not represent every segment equally.
10. Platform behavior can change after the September 2026 endpoint.
11. The exploratory intervals do not establish causality.
12. No downstream financial validation has yet been completed.

What We Will Test Next

The current signals should remain frozen and be compared with outcomes that become observable later.

For healthcare companies, the most useful validation sequence is likely:

1. Branded search and digital demand: Does recommendation momentum precede changes in branded search, website traffic, plan-shopping activity, pharmacy engagement, or direct-to-consumer testing interest?
2. Operational outcomes: Do later membership, prescription-related, diagnostic-testing, enrollment, or segment-level demand measures move in the same direction where comparable public data exists?
3. Analyst expectations: Does AI momentum precede revenue-estimate revisions after controlling for information already available at the signal date?
4. Reported financial performance: Does the AI-enhanced model improve prediction of later revenue growth or revenue surprise beyond a conventional baseline?
5. Market outcomes: Only after earlier commercial links are established should later sector-relative or factor-adjusted stock-return tests be interpreted.

Any of these relationships may prove weak, conditional, or nonexistent.

That possibility is part of the research design, not a failure of publication discipline. The falsification criteria are published in What Would Prove the AI Commercial Momentum Hypothesis Wrong?.

Related LLM Authority Index Research

• Can AI Search Signal Future Revenue Growth? The AI Commercial Momentum Hypothesis
• Initial Findings From 25 Public Companies
• How We Measure AI Commercial Momentum
• AI Investor Signal Tracker
• AI Recommendations vs. Mentions vs. Citations
• AI Recommendation Share vs. Market Share
• Does Cross-Platform AI Visibility Matter?
• Insurance Stocks and AI Search
• Molina Healthcare AI Search Visibility
• Cigna AI Search Visibility
• UnitedHealth Group AI Search Visibility
• CVS Health AI Search Visibility
• Labcorp AI Search Visibility
• The Persistence-Portability Gap

Research Status Reminder

Research status: Exploratory longitudinal research. The current AI recommendation signals have not been validated as predictors of revenue, earnings, analyst revisions, valuation, or stock returns.