July 13, 2026 U.S. hospitals topic labeling

Patient Experience Through Hospital Reviews

What people value in U.S. hospitals, where friction persists, and how hospital teams can set priorities.

Google reviews of U.S. hospitals and other healthcare facilities describe both positive clinical interactions and frustrating processes. Bedside manner is a frequent source of praise, waiting and reception account for a substantial share of negative feedback, and financial concerns remain overwhelmingly negative even as the overall tone improves.

For a hospital team, those findings call for different responses: preserve what people value, give recurring friction an operational owner, and keep persistent or serious concerns visible in their own right. A single average cannot do all three jobs.

The analysis follows 217,343 Google reviews of 10,029 U.S. healthcare listings, from 2019 to early June 2026. Topic findings use 196,183 labeled accounts. They describe a self-selected public record. Tone measures below use positive-or-negative statements; neutral ones are excluded.

Protect the strengths; distinguish the problems

Clinical bedside manner appears in 41.7% of labeled accounts, waiting in 26.2%, and front-desk conduct in 25.3%. These are the most frequently discussed topics, but their frequency does not tell us which response each needs.

The chart separates two questions: how often a topic appears, and how negatively it is described when it does. Together, they distinguish a widely valued aspect of the experience from recurring friction and smaller, consistently negative issues.

The most common topics are not the most consistently negative Each dot is one of the 29 topics. Select a topic for its exact values, or hover a dot. The dashed line marks the overall negative share among directional statements, 45.6%.

Source: saved topic reach and sentiment aggregates. A review can mention several topics, so reach percentages do not add to 100%. Neutral statements are excluded from negative-share calculations.

Preserve clinical strengths. Only 20.9% of directional bedside-manner statements are negative. For a service lead, this is a reason to collect and share checked examples of listening, respect, and compassion in team learning and onboarding. It also leaves substantial negative feedback to understand: a common strength can still contain important exceptions.

Give recurring friction an operational owner. Waiting is the largest source of negative statements in the 29-topic framework, with 29,692. Together, waiting and front desk account for 25.8% of all negative statements. Front-desk tone is mostly positive, yet its large reach produces considerable negative volume. Put access and registration alongside clinical interactions on the patient-experience agenda, then identify the particular local process that needs to change.

Keep persistent and serious concerns visible. Billing appears in 4.6% of labeled accounts, but 93.6% of its directional statements are negative. Patient safety, privacy, and equitable-treatment topics also combine lower reach with strongly negative tone. A smaller volume is a reason for focused follow-up, rather than a reason to let an issue disappear behind the larger topics. Reports of harm or unfair treatment need human assessment through the hospital’s existing processes.

The practical priority is therefore not a single ranked list. Ask how much negative feedback an issue generates, how concentrated that negativity is, and whether the account describes something serious. Keep the strengths in view while deciding what to improve. These are complementary questions, not a formula for predicting the benefit of an intervention.

Explore every topic and the underlying taxonomy

The taxonomy article explains how the 29 topics were discovered and refined. The original visual report retains the topic-frequency bars, category treemap, and unadjusted rating comparisons. For the separate question of how topics relate to ratings after accounting for other recorded features, see Which Experiences Are Linked to Lower Hospital Ratings?.

Feedback is becoming more positive, but financial concerns persist

The overall negative share falls from 52.4% in 2022 to 40.4% in 2025, a decline of about 12 percentage points. Mean star ratings rise over the same period. That is an encouraging change in the public record; the useful next question is which parts of the experience account for it.

Follow the record over time Quarterly observations, 2019 to early June 2026. The final quarter is incomplete. Sentiment and ratings come from related accounts and are not independent validation of one another.

Source: saved quarterly aggregates. The count series contains 217,155 sampled reviews with dates in the plotted window; rating averages use dated sampled reviews, and negative share uses directional labeled statements. The dashed final segment leads to the incomplete quarter.

Waiting and front-desk topics account for 5.88 percentage points, or 49.2% of the decline. The contribution chart shows where the aggregate change sits, rather than treating every topic as if it moved equally.

Waiting and front-desk topics account for about half of the decline Eight largest reductions in contributions to the overall negative share, 2022 to 2025. Waiting contributes −3.55 percentage points and front-desk conduct −2.33, together about 5.9 of the roughly 12-point decline.

Source: saved contribution calculation. Only the eight largest reductions are plotted; they do not sum to the entire change. This is an accounting breakdown, not a causal attribution.

The change is also visible within those topics. Waiting statements fall from 61.6% negative in 2022 to 44.9% in 2025; front-desk statements from 45.1% to 29.4%. The picture is more specific than “ratings went up”: accounts of waiting and reception became less negative within their own topic categories.

A topic’s contribution can also fall because it occupies less of the conversation. The sample’s listings, contributors, and services can change too. These figures describe feedback; they do not establish shorter measured waits or isolate the effect of a hospital intervention.

The financial experience needs its own workstream

The annual heatmap exposes what the improving aggregate can hide. Financial & Administrative feedback remains above 90% negative in every complete year from 2019 to 2025. It falls from 94.0% in 2022 to 91.0% in 2025, while Staff, Communication & Nursing falls from 37.2% to 25.9%.

Finance has become somewhat less negative, but it remains a persistent concern. Improvement elsewhere should not close the discussion of charges, coverage, or the experience of getting an issue resolved.

The improving aggregate contains very different topic trends Negative share of directional statements by topic group and year. Select a year for exact values; 2026 is partial through early June.

Source: saved annual topic-group aggregates. The financial group contains billing and insurance topics. Colors describe tone, not clinical severity.

Give financial experience a named owner and follow-up measures of its own. A team could separate charge disputes, coverage questions, and explanation or response problems, choose a locally supported process change, and track resolution time and repeat contacts. The action follows from the persistence of the signal; the particular fix must follow from the hospital’s cases.

Supporting check: the trend within listing cohorts

The longest-observed cohort also becomes less negative: 53.9% in 2022 to 42.1% in 2025. That makes newly entering listings an incomplete explanation of the aggregate change. Cohorts are defined by their first observed review year; membership within each year’s data, services, contributors, and topic mix can still change.

Inspect the record by listing cohort Groups describe the first observed review year, rather than when a hospital opened. 2026 covers the period through early June.

Source: saved cohort-year aggregates; cells require at least 300 directional statements. This is not a fixed-hospital estimate of improvement in care.

A star average compresses experiences that need different responses

The sample’s 3.54-star average combines a polarized distribution: 89.2% of accounts give either one or five stars. An average is useful for tracking a summary, but this distribution tells a hospital team that the public record is dominated by strongly favorable and strongly unfavorable assessments.

One-star and five-star ratings account for 89.2% of reviews Share of the 217,343 sampled reviews at each rating. These are reviews, not extracted statements.

Source: saved rating distribution; percentages sum to 100% before rounding.

An overall assessment also does not describe every part of an account. In the labeled subset, 14,639 accounts (7.5%) contain both a positive and a negative topic statement. Among 64,283 labeled one-star accounts, 3,951 (6.1%) contain a positive topic statement.

Mixed feedback is a minority, but a measurable one. A low rating can identify a problem to resolve and a behavior to preserve. When learning from an account, separate those two contributions instead of treating every sentence as a complaint.

For reporting, show the rating distribution alongside the mean. For service recovery, identify the issue requiring a response while retaining any specific praise. For tracking change, use the same definitions and time window: a rising mean alone does not identify which process changed.

Choose the right comparison before judging performance

Psychiatric-hospital listings have approximately 72% negative tone, compared with 28% for urgent-care listings and 53% for emergency-room listings. These are different reference groups. A national or state average that combines them may be a poor comparison for a particular service.

The tone differs substantially across listing categories Negative share of directional statements. Categories require at least 1,000 directional statements; the dashed line is the overall 45.6% share. These are unadjusted comparisons.

Source: saved Google listing-category aggregates. Categories are platform labels and can misdescribe a setting; statement counts do not measure patient volume.

Start with comparable services and the same period, showing the volume and composition of feedback alongside tone. The category differences demonstrate why context matters; they do not establish clinical-quality differences or explain how much of a state’s overall result comes from its service mix.

Where the sample comes from

The coverage map shows the source of that context. The collection includes hospitals, emergency departments, clinics, and urgent-care centers; Google listings do not necessarily correspond one-to-one with physical hospitals. California, Texas, and Florida supply 39.2% of sampled accounts.

Geographic coverage across the United States217,135 reviews matched to a US state
Hospital review coverage by US state States are shaded by the selected measure. Facility locations can be shown as a dot overlay.

Source: saved sample coverage and listing coordinates. 217,135 reviews are matched to a U.S. state; 208 are not. Counts use a square-root color scale; average ratings use a linear scale.

The tone map answers a different question: how the collected statements are described across states. It displays the 40 states with at least 1,500 directional statements. Gray states fall below that threshold. Use the count view to inspect the volume behind a comparison.

Where reviews run hot40 states meet the 1,500-statement threshold
Negative share of hospital review statements by US state States meeting the publication threshold are shaded by the selected measure. Facility locations can be shown as a dot overlay.

Source: saved state aggregates. Negative share uses negative ÷ (negative + positive) statements. The share uses a linear color scale; statement counts use a square-root scale. Location dots show the broader sample, including states below the sentiment threshold.

State estimates range from roughly 36% negative in Florida, Texas, and Nevada to 62% in New York. Read that spread alongside the care-setting chart and coverage map, rather than as a hospital-quality league table. A formal performance benchmark would need to account for relevant differences in services, populations, and the reviewed record.

The useful local question is whether a pattern remains when you compare the relevant service with an appropriate peer group. The maps help locate that question; the category chart helps define the comparison. The original city comparison remains available as supporting context.

Turn the findings into a small number of owned improvements

The findings support a practical agenda: preserve valued staff behaviors, give recurring access problems an operational response, and manage persistent financial concerns separately. Serious reported concerns should continue through their own human assessment route.

The table below translates that agenda into proposed next steps. The measures are examples to use locally; this study did not evaluate these interventions or their results.

Workstream and possible owner Next step and evidence of progress
Access and registration
Operations/access and registration leads, with patient experience		 Select one recurring wait or registration problem. Distinguish throughput, administrative completion, and interaction quality before choosing a change. Track: the relevant wait-stage or contact-completion measure, alongside topic-specific feedback over the same period.
Financial experience
Patient financial services and coverage/authorization team		 Clarify ownership and the resolution route for charge disputes and coverage questions. Choose a locally supported process change. Track: resolution time, repeat contacts, recurring issue types, and financial-topic feedback.
Safety, privacy, and equitable treatment
The hospital’s existing responsible teams		 Retain a separate human assessment and escalation route, including lower-volume concerns. Track: validated case follow-up; assess seriousness from the account and appropriate evidence, independently of its star rating or topic volume.
Strengths in staff interactions
Relevant service/nursing leads and patient experience		 Use checked examples of valued behaviors in team learning and onboarding, and preserve them during process changes. Track: feedback about those behaviors alongside appropriate local staff and patient evidence.

For the next improvement discussion, bring one recurring access issue and one persistent financial issue, each with an owner, a proposed change, and a measure that would show whether it helped. Choose the particular issues from the hospital’s own cases and service context. Keep the positive behaviors visible and serious concerns on their existing escalation route.

The reviewer-composition analysis examines whether contributor mix materially changes ratings. The topic-pair analysis identifies issues worth reading together, and the rating-association analysis asks which mentions remain associated with ratings after adjustment. Here, the useful outcome is an agenda with distinct questions and owners, rather than a single experience score expected to carry every decision.

Supporting detail: what the source material contains

The median sampled account contains 37 words, with a long tail of more detailed accounts. The length chart is retained here as context for the material being interpreted. Length itself does not establish seriousness or sentiment.

Most reviews are short, with a long tail

The median sampled review has 37 words. This density plot preserves the original distribution and its 2,500 sampled observation dots. The longest 1% of reviews, those exceeding 394 words, are omitted from this view. The vertical positions of the dots are jittered for visibility and do not encode sentiment or ratings.

Source: saved density estimate and sampled points. The dashed line marks the 37-word median. The shape describes length; it does not by itself show that long reviews are complaints.

Definitions, sources, and all original visualizations

Units. A review has one star rating and can contribute several labeled statements. Topic reach is the percentage of the 196,183 labeled reviews mentioning a topic at least once. Negative share is negative ÷ (negative + positive) statements. Negative statement volume counts label rows, rather than patients or care events. A statement has one topic in this accounting, so topic counts partition the 192,195 negative statements; reviews can mention several topics. Neutral statements are included in the 425,667 total but excluded from tone calculations.

Scope and sampling. The coverage sample was drawn from about 2.09 million eligible text reviews and contains 217,343 reviews of 10,029 U.S. healthcare listings, spanning 2019 to early June 2026. Sampling within listing and year, capped at 200 per listing, limits the influence of heavily reviewed locations. The 20-review minimum is source-listing eligibility, not a guarantee of 20 sampled or labeled accounts in every comparison. Unweighted results describe the sample. They do not represent all patients, care delivered, or total Google reviewing volume. Valid labels cover a narrower 9,904 listings. Collection ends in early June; 2026 is incomplete.

Label quality. Automated checks verify permitted topic names and IDs, quoted phrases, review resolution, and known sentiment. The saved report’s 98.8% mechanical validity rate is not human labeling accuracy. About 9.7% of sampled accounts have no retained statement; this summary does not establish why each was unlabelled. Serious labels describe reported concerns, rather than verified incidents.

New narrative calculations. Negative statement counts, within-topic 2022/2025 rates, and mixed-feedback counts were recomputed from the frozen label parquet using the same validity gates and distinct listing-plus-review keys. All 29 topic reach/tone values match the original chart snapshot. Mixed accounts have at least one valid positive and one negative statement; the one-star-with-praise figure requires at least one positive statement in a labeled one-star account. The annual contribution changes also match the saved chart data. Download the narrative evidence.

Trends and comparisons. A contribution divides a topic’s negative statement count by all directional statements that year. Its change combines topic mix and within-topic tone, with no adjustment for changing listings, services, or contributors. Listing cohorts are defined by first observed review year, not hospital opening date, and do not hold individual listings constant. State and category comparisons are unadjusted; platform categories can misdescribe services. Neither coefficients nor expected intervention benefits are estimated here.

Preserved figures and reproduction. Every original chart view, both maps, the full topic table, and their controls remain available on this page. The original exploratory report retains the additional taxonomy, treemap, city comparisons, topic bars, and rating gaps. Its stronger causal and care-quality interpretations are superseded by the wording here. Download the chart aggregates, map data, or new narrative evidence. The existing generator extracts saved chart outputs; the new evidence generator recomputes aggregates from frozen labels. Neither reruns a model or exports review text or contributor identifiers.