Healthcare generates more data than ever, yet the patients and providers who need it most are routinely making decisions based on an incomplete picture. Fragmented records, inconsistent standards, and siloed systems mean that the average patient’s health history is spread across a dozen or more platforms, each capturing a different slice of the story.
As health plans and health systems race to deploy AI-powered tools — across telehealth and virtual care, chronic condition management, AI health assistants, and behavioral health — it’s the quality of the underlying data that determines whether those tools help or harm.
This report makes the case for shifting the industry’s focus from data coverage to data completeness and explains how b.well Connected Health has the infrastructure to make the shift.
Key Findings
- 83% of healthcare consumers say it is critical that AI works with all their doctors, pharmacies, and insurance providers.
- 85% of those with chronic conditions want AI that enables care coordination across providers
- Digital health services depend on the same complete record, whether the interaction is a telehealth visit, a wellness nudge, a condition management program, or a behavioral health session.
- Raw patient data can contain hundreds of medication and lab records that reduce to just a handful of clinically relevant entries when properly refined.
- Unrefined data leads to AI hallucinations, dangerous drug interaction misses, duplicated diagnostics, and 10x higher token costs.
- b.well’s 13-step Health Data Refinery transforms fragmented, multi-source healthcare data into complete, AI-ready patient records.
Table of Contents:
- The Problem: Healthcare Data Is Broken
- What Patients Are Telling Us
- The Critical Distinction Between Coverage & Completeness
- What Raw Data Actually Looks Like
- The b.well Health Data Refinery
- b.well’s Connectivity Infrastructure
- Implications Across Digital Health and the Broader Ecosystem
- The Path Forward
1. The Problem: Healthcare Data Is Broken
Healthcare sits in a paradox where clinicians and health plan administrators have more data at their disposal than at any point in history, yet the data they can actually act on is incomplete.
The average patient’s health information lives across dozens of systems: electronic medical records (EMRs), pharmacy dispensing platforms, insurance claims databases, laboratory information systems, health information exchanges (HIEs), and consumer health apps. Each system was built independently, speaks a different technical dialect, and captures a different portion of the clinical picture.
The consequences are well-documented and serious:
Incomplete medication lists lead to harmful drug-to-drug interactions.
Missing allergy records result in preventable adverse reactions.
Fragmented care histories lead to duplicate testing and delayed diagnoses.
Siloed specialist notes mean primary care physicians make decisions without full context.
Outdated insurance records drive coverage denials and billing errors.
For health plans, this fragmentation undermines the effectiveness of care management, drives up medical costs, and creates failures in member experience. For health systems, it means clinicians are practicing with incomplete information at the point of care. For digital health companies, it means a telehealth visit starts with a blank intake form, a condition management program guesses at what a member needs next, and a behavioral health provider has no view of the physical health picture shaping their patient’s care.
The arrival of AI compounds the stakes. As organizations invest in AI for care gap closure, predictive risk stratification, clinical decision support, and member navigation, the quality and safety of their outputs is directly dependent on the quality of the data feeding those models. Artificial intelligence is only as good as the data it learns from. Feed AI fragmented, inconsistent, low-quality data, and you will get unreliable, potentially harmful results.
2. What Patients Are Telling Us
Consumer adoption of AI for health is accelerating faster than the healthcare system can track. Rock Health’s 2025 Consumer Adoption Survey, fielded in December 2025 across 8,000 U.S. Census-matched adults, found that one in three (32%) U.S. adults has used an AI chatbot for health information — double the share from just a year prior. Of those users, 64% engage with AI for health questions weekly or more, and 81% report taking at least one action as a direct result of their AI interaction, including 40% who consulted a provider and 18% who adjusted their medications.
This is not passive browsing. Consumers are using AI to make real health decisions. And they are largely doing so with general-purpose tools: nearly three-quarters of AI health users reported using ChatGPT, compared to only 5% who used a provider-offered chatbot and 4% who used a payer-offered chatbot. Demand for always-on, personalized health guidance was already pent-up before purpose-built healthcare AI experiences arrived. The question is no longer whether consumers will use AI for health — it is whether the AI they are using has access to the data it needs to be safe.
A survey of 598 healthcare consumers conducted in early 2026 revealed a finding that should recalibrate how health plans and health systems think about AI adoption: the people who need AI-assisted healthcare most are also the most skeptical of it.
Among respondents managing chronic conditions, only 47% trust AI for health information, compared to 52% of healthy individuals. 78% of those with chronic conditions say a doctor or nurse must review what AI tells them, compared to 73% of people without chronic conditions.
People managing diabetes, hypertension, kidney disease, or multiple concurrent conditions have experienced firsthand what incomplete data looks like in practice: the specialist who did not know about the drug allergy, the medication interaction that was not caught, the test repeated because results did not transfer, the contradictory treatment plans from two providers who never communicated, and therefore have a right to be skeptical.
What Patients Actually Want from Health AI
When asked to rank the features that matter most in AI healthcare tools, consumers with chronic conditions were unambiguous:
| 85% | Want AI to help their doctors work together better |
| 83% | Say it’s critical that AI works with ALL their doctors, pharmacies, and insurance |
| 82% | Want all their health information brought together from different sources |
| 79% | Want test results explained in context, not in isolation |
| 91% | Want to require that AI has been tested and proven accurate |
These are articulations of pain points that patients live with every day, and a clear mandate for what health plans and health systems must deliver for AI adoption to succeed.
Notably, 65% of respondents prefer AI integrated into their patient portal over standalone AI chatbots. Patients want AI that already has their data unified. The implication for health plans and health systems is that AI deployed on top of siloed data will not earn members’ or patients’ trust, regardless of how sophisticated the underlying model is.
The 60% Majority: Your Core Market
Sixty percent of survey respondents report managing ongoing health conditions. This is your highest-utilization, highest-cost, highest-need population. These individuals are already motivated to engage with health technology: 76% have searched Google for health information in the past three months, and 85% want proactive AI features like monitoring, explanations, and care coordination.
The gap is not in demand; it is in trust, and trust can be earned through completeness.
3. The Critical Distinction Between Coverage & Completeness
For years, the healthcare industry has measured data strategy success by coverage — how many sources are connected, how many patients are represented, and how many records have been ingested. Coverage is necessary, but it is not sufficient.
Consider what this actually delivers in practice. A health plan connects to a regional hospital system’s EMR and ingests its patient records. Those records show recent diagnoses, some medications, and recent lab values. Coverage box: checked. But what about the years that came before, including:
Coverage captures what a single system knows, but completeness captures the full clinical reality of a person’s longitudinal health. The massive gap between those two things is where AI fails, and patients get hurt.
Three Dimensions of Completeness
True data completeness requires three things working together:
Depth: Longitudinal History, Not Snapshots
A single cholesterol reading is nearly meaningless in isolation. Paired with readings from the prior 18 months, medication titrations, and correlated lab work, it becomes a trend that can drive clinical action. 79% of consumers want AI to explain test results in context. This requires longitudinal data that show how health metrics evolve over time, how treatments are adjusted, and how interventions correlate with outcomes.
Coverage: All Sources, Not One System
The average person with a chronic condition sees multiple specialists, uses different pharmacies, has held several insurance plans, and has received care across various facilities. Their complete health story may live in 10 or more systems. AI that accesses only one slice of this picture is not providing clinical intelligence. It is providing a partial view that, presented with confidence, can be more dangerous than no view at all.
Currency: Updated, Accurate, and Verified
Depth and breadth are meaningless if the data is stale. Anyone who has reviewed a patient portal knows the problem: medications stopped two years ago still appear as active, conditions that have been resolved remain on the list, and outdated contact information persists alongside correct information. The data exists, but it has not been maintained. Complete data is not just comprehensive, it is current.
4. What Raw Data Actually Looks Like
An analysis of millions of real, anonymized patient records shows the stark difference between raw healthcare data and properly refined records.
The table below highlights what a typical patient record looks like before and after b.well’s refinement process, and what the differences mean for AI performance and safety.
The aggregate impact of unrefined data on AI systems causes responses to run 10x slower, token costs to run 10x higher, and the clinical outputs to be unreliable. Properly refined data inverts all of these outcomes.
One real-world example clearly illustrates the risk. A patient with diabetes and hypertension has their blood glucose managed by a primary care physician and their cardiovascular risk monitored by a cardiologist at different health systems. An AI tool with access only to the PCP’s records recommends a medication adjustment. It does not know about the cardiologist’s recent prescription, which creates a contraindicated interaction. With complete, cross-system data, the interaction is flagged. Without it, the recommendation is confidently and incorrectly delivered.
Independent research from Google confirms what the data above illustrates. In Google’s Personal Health Record (PHR) research, published in May 2026, researchers evaluated the impact of providing AI models with access to a complete personal health record versus operating without that context. The results were unambiguous: both independent clinicians and AI raters judged AI responses to be significantly more helpful when the model had access to complete PHR data. In a separate study, Google’s Plan for Care research found that 15% more participants felt better prepared for their doctor’s visit, and 13% more felt confident they could make the most of their appointment, when AI had access to their complete health record context.
These improvements are the direct result of giving AI complete information rather than a fragment of it. The infrastructure question — whether the data feeding an AI model is complete, normalized, and current — is not a backend concern. It is the determinant of whether AI helps or misleads the patient using it.
5. The b.well Health Data Refinery
b.well Connected Health has spent more than a decade building the infrastructure platform to solve the completeness problem. At the core of the platform is a 13-step Health Data Refinery which is a systematic process that transforms fragmented, multi-format, multi-source healthcare data into unified, clinically accurate, and AI-ready longitudinal records.
The 13-Step Refinery Process
Each step addresses a specific class of data quality problem that, left unresolved, propagates errors into downstream AI and clinical applications.
1. Data Conversion
Ingests data in any format—X12 claims, HL7 messages, C-CDA documents, CSV files, JSON APIs—and converts everything to standardized FHIR resources with a common language across all sources.
2. Data Cleansing
Identifies and corrects errors: standardizing date formats, fixing medication name typos, correcting invalid codes, and removing obviously erroneous values.
3. Data Quality Assurance
Evaluates incoming data against quality thresholds. Rejects records that fail minimum standards; questionable data is flagged for review rather than silently accepted.
4. Data Lineage
Maintains complete transparency about data origins. Every piece of information carries metadata showing its source, receipt timestamp, and transformation history—creating an audit trail for compliance and trust.
5. Data Standardization
Names, addresses, phone numbers, medication codes (RxNorm), and diagnosis codes (ICD-10) are all normalized to consistent formats, enabling reliable cross-source matching and analysis.
6. Data Normalization
Identifies and investigates outliers: a prescription for 10,000 pills, unusual dosages, extreme lab values. These are validated against clinical guidelines before acceptance.
7. Data Enrichment
Enhances raw data with consumer-friendly descriptions, common side effects and interaction information, relevant educational content, cost and coverage information, and alternative treatment context.
8. Data Linking (Patient Matching)
Uses probabilistic matching algorithms to link records from different sources to the same patient—even when names differ slightly, addresses have changed, or demographic data is inconsistent. This is the step that prevents a patient’s records from being treated as those of two separate people.
9. Data Event Linking
Connects related records about the same clinical event. The EMR prescription record, the pharmacy fill record, and the insurance claim are linked as different perspectives of a single event, rather than being treated as independent data points.
10. Data Merging and Deduplication
Intelligently merges linked records, combining dosage instructions from the EMR, fill dates from the pharmacy, and costs from insurance into one comprehensive medication record. Duplicates are identified and consolidated. This step reduces 700+ raw medication records to 29 clinically relevant ones.
11. Data Gap Identification
Identifies what’s missing: incomplete medication histories, absent lab results, overdue preventive screenings. Prompts patients to connect additional data sources or fill in gaps, turning passive data aggregation into an active completeness process.
12. Data Interpretation
Rules engines and machine learning models add clinical intelligence: identifying condition categories, calculating risk scores (including HCC codes), flagging potential medication interactions, and detecting care gaps and quality measure opportunities.
13. AI Optimization
Prepares data specifically for AI consumption: generating International Patient Summaries, extracting and structuring information from clinical notes, creating temporal event timelines, and building knowledge graphs that represent complex relationships between conditions, medications, and outcomes.
6. b.well’s Connectivity Infrastructure
The refinery process is only as powerful as the breadth of data flowing into it. b.well has built one of the most connected health data platforms in the industry, with integrations spanning every major source of patient health information:
| Data Source | Coverage |
|---|---|
| Providers | Nearly 2.4M provider connections, enabling real-time access to medical records, lab results, imaging reports, and clinical notes |
| Pharmacy Networks | National coverage, including major chains and mail-order pharmacies, for complete medication histories and real-time prescription data |
| Insurance Payers | All major payers for eligibility, benefits, claims, and coverage information |
| Wearables & Health Apps | Apple Health, Fitbit, and other consumer devices for continuous monitoring of data |
| Laboratory Networks | Direct connections to LabCorp and independent facilities for test results |
| Social Determinants | SDOH data to contextualize the full scope of a person’s health journey |
Breadth of connectivity is essential, but it is not the whole story. b.well normalizes data across different formats and standards, reconciles duplicates and conflicts, and maintains longitudinal histories that show how health evolves over time. The result is not just more data, it is better data.
7. Implications Across Digital Health and the Broader Ecosystem
For Health Plans
Data completeness is the precondition for effective care management. When a member’s record includes only their in-network claims history, care gap programs miss interventions addressed by out-of-network providers. Risk stratification models trained on incomplete data misclassify members, directing resources to the wrong populations. AI-powered member navigation tools give incorrect or dangerous answers about coverage, medications, or care options.
Complete data enables health plans to close the gap between what they know and what is clinically true for each member, enabling more precise risk stratification, more effective gap-in-care programs, and AI-assisted engagement that members will actually trust.
The survey data is instructive here: 65% of consumers prefer AI integrated into their patient portal vs. standalone tools. Members are not opposed to AI assistance. They just want it to be comprehensive. Health plans that deploy AI on top of siloed data will find that the 60% managing chronic conditions, their highest-cost population, disengage.
For Health Systems
At the point of care, incomplete data creates avoidable risk. Clinicians relying on AI for clinical decision support, diagnostic assistance, or treatment recommendations need confidence that those recommendations reflect the full patient picture, not just what happened within their own system.
When a patient presents with a complex chronic condition managed across multiple specialists, the AI that supports their care must have access to all of it: the specialist notes, the out-of-network lab work, the prescriptions filled at the corner pharmacy, the wearable data showing how blood pressure responds to activity. Without that, the AI is not a clinical intelligence tool, it is a liability.
b.well’s refinery enables health systems to provide clinicians with a complete longitudinal view of patients, regardless of where care was received. This reduces duplicated testing, prevents interaction errors, and enables the kind of AI assistance that earns clinician trust.
For Digital Health
Digital health now spans a wide range of services: telehealth and virtual care, fitness and wellness apps, chronic condition management platforms, AI health assistants, and behavioral health providers. What they share is a dependence on complete, consented health data to deliver clinically meaningful experiences. A telehealth visit that opens with a blank intake form repeats questions the record already answers. An AI-powered diabetes app that can only see what a user manually enters is fundamentally limited. A behavioral health provider without visibility into medications and physical conditions is treating half the person. Access to longitudinal clinical records, pharmacy data, lab results, and wearable readings is what turns each of these into personalized, contextual guidance that drives real behavior change and measurable outcomes.
b.well’s modular SDKs and APIs enable digital health builders to embed complete health data into their products without rebuilding the underlying data infrastructure—connecting to nearly 2.4 million providers, 350+ health plans, labs, pharmacies, and device networks through a single integration. The result is AI that can access consented, longitudinal health data and produce clinically grounded, personalized insights rather than generic responses.
For Life Sciences and Pharma
Life sciences and pharmaceutical organizations face a persistent challenge: accessing diverse, real-world patient populations with the data quality and consent infrastructure required for rigorous research. Incomplete or fragmented data slows trial recruitment, undermines real-world evidence generation, and limits the ability to monitor therapy adherence at scale.
b.well addresses each of these challenges through longitudinal FHIR-based health records, a consent-by-design framework that supports HIPAA, marketing, and research compliance, and automated workflows for patient recruitment, digital data capture, and therapy companion experiences. Organizations can recruit smarter and more inclusively by reaching underrepresented and diverse populations through digital channels, collect patient-reported outcomes with less friction, and support therapy adherence through personalized digital experiences—all built on a data foundation that is modern, complete, and compliant.
For Patients and Clinicians: Extending Clinical Judgment, Not Replacing It
75% percent of all survey respondents, and 78% of those with chronic conditions, want a doctor or nurse to review what AI tells them. This is not a barrier to adoption. It is a design requirement.
b.well builds human oversight into the core architecture of its platform. AI can draft messages to clinicians, but patients review and send them. It can suggest appointments, but patients confirm them. It can surface clinical context, but it consistently directs users to follow up with their care team for medical decisions.
The role of AI in healthcare is to synthesize complete data at a scale no human can match, and then put that synthesis in the hands of patients and clinicians who can act on it. Replacing clinical judgment is neither the goal nor the right outcome. Extending it with complete information is.
8. The Path Forward
Healthcare organizations are moving fast to deploy AI across care management, clinical decision support, and member engagement, and for good reasons. The potential for better outcomes, lower costs, and improved experiences is real. But speed without the right foundation is a familiar mistake, and the data foundation most organizations are building on is broken.
The survey data, the real-world analysis of patient records, and the experience of millions of patients managing complex conditions all point to the same conclusion: data completeness is the prerequisite.
Four Non-Negotiables for Trustworthy Health AI
1. Interoperability First, AI Second
Before investing in algorithms, invest in data infrastructure. FHIR APIs, health information exchanges, patient-mediated data sharing, and multi-payer connectivity are not nice-to-haves. They are the prerequisite for AI that delivers reliable value. 80% of consumers say AI must work with all their providers. Earning their trust depends on it.
2. Completeness Over Coverage
Measure your data strategy by the clinical completeness of individual patient records, not by the number of connected sources. The right data, refined and reconciled, is what drives reliable AI.
3. Transparency About Data State
82% percent of consumers want control over what data AI uses. Show users what sources are connected, flag what is missing, and give them the ability to fill gaps. Transparent AI built on complete data will earn trust.
4. Build for Complex Cases
Build for the most complex cases first. If AI works reliably for a patient managing diabetes, hypertension, and kidney disease across four specialists and two health systems, it will work for everyone. Optimizing only for simplicity means failing the 60% who have ongoing health conditions.
Conclusion
The future of healthcare AI is not determined by which organization deploys the most sophisticated model. It is determined by which organizations build on the most complete data.
Health plans, health systems, and digital health companies that invest in data completeness, not just data coverage, will deploy AI that their highest-need populations actually use and trust. They will close care gaps with precision. They will enable clinicians with insights that reflect the full clinical picture. And, they will earn the trust of the 34% of consumers now waiting for AI to earn it.
b.well has spent more than a decade building the infrastructure to make this possible. Our Health Data Refinery, our connectivity across 300+ health systems, national pharmacy networks, all major payers, and leading lab networks, and our commitment to human-centered design presents the best path to conquer the completeness challenge.
Because in healthcare, gaps in data don’t just slow things down, they lead to real harm. And the patients we are all trying to help understand this better than anyone.
About the Research
Consumer survey data cited in this white paper is based on a study of 598 healthcare consumers conducted in January–February 2026, examining attitudes toward AI in healthcare, trust factors, feature preferences, and barriers to adoption. Respondents spanned all age groups (18–65+), with approximately 60% reporting ongoing health conditions they manage regularly. Data completeness analysis is based on b.well’s analysis of millions of real, anonymized patient records
Additional Sources
Rock Health. “The tortoise and the hare of care: Health AI insights from Rock Health’s 2025 Consumer Adoption Survey.” March 23, 2026. rockhealth.com
Google Research. “A New Era of Discovery: Google Research at I/O 2026.” May 28, 2026. research.google
About b.well Connected Health
b.well Connected Health is the most data-rich digital health platform for AI-powered consumer experiences, personalized care, and better outcomes. The company solves healthcare’s fragmentation problem with a scalable, FHIR®-based platform that unifies health data, solutions, and services in one place. By creating longitudinal health records, cleansing and standardizing fragmented data through its proprietary Data Refinery, and delivering proactive insights, b.well empowers organizations to engage consumers in real time, simplify access to care, and support regulatory compliance. Learn more at www.bwell.com.
