AI in Primary Care: What's Hype, What's Real, and What Saves Time

How to Think About AI Claims in Healthcare

Every EMR vendor, health system, and technology startup has an AI story in 2026. The claims range from "faster note generation" (measurable, testable) to "improves clinical decision quality" (complex, context-dependent) to "transforms patient outcomes" (unmeasured, aspirational). As a physician evaluating these claims, you need a framework for separating evidence-based AI from marketing-grade AI.

Three questions to ask about any AI claim:

1. Is this published in a peer-reviewed journal, or is it from a vendor white paper?
2. What is the control condition? (AI vs. no AI, or AI vs. current best practice?)
3. Was this studied in a setting similar to mine (primary care, outpatient, similar volume)?

What Works: AI Applications with Published Evidence

1. Ambient AI Documentation (Strong Evidence)

This is the most well-evidenced AI application in outpatient primary care. Multiple peer-reviewed studies (JAMA Network Open 2023, NEJM Catalyst 2024, Health Affairs 2024) demonstrate:

• Documentation time reduction: 25–50% per encounter
• After-hours documentation decrease: 30–40%
• Physician-reported note quality: equivalent or better than self-authored notes in 80–90% of encounters
• Burnout score improvement: 15–25% on validated instruments in 6-month studies

What to look for: The AI listens passively during the encounter and produces a draft note that the physician reviews and signs. No dictation script, no command words, no changing how you speak to patients. Systems that require the physician to alter their behavior to serve the AI reduce adoption and limit benefit.

2. AI-Assisted Coding and Billing Review (Moderate-Strong Evidence)

AI that analyzes note content and suggests ICD-10 and CPT codes, flags potential undercoding, and identifies missing HCC-relevant diagnoses has demonstrated revenue impact in multiple health system studies:

• A 2024 University of Michigan study found 8–12% additional annual revenue per physician from AI coding review
• Claim denial rate reductions of 30–50% when AI flags documentation gaps before submission
• HCC capture improvements of 15–25% for Medicare Advantage panels

3. Medication Interaction and Dosing Alerts (Well-Established)

Rule-based drug interaction checking has been in EMRs for 20+ years and reduces serious drug interactions. The challenge is alert fatigue: physicians override 90%+ of alerts in systems with poorly calibrated rules. AI-enhanced alerting that learns which alerts a given physician acts on and suppresses those routinely overridden is associated in published literature with substantially lower override rates—because the alerts that remain are the ones the physician actually considers.

4. Sepsis and Deterioration Early Warning (Hospital Setting; Limited Outpatient Evidence)

AI-based early warning systems (Epic's Deterioration Index, Rothman Index) have demonstrated reduced sepsis mortality in inpatient settings. The outpatient corollary—identifying patients at risk for deterioration before an emergency visit—is actively studied but does not yet have the same evidence base. If an outpatient AI vendor claims their system prevents hospitalizations or emergency visits, ask for the study design and control condition before accepting the claim.

What Remains Aspirational: Where Evidence Is Weak or Absent

AI-Assisted Differential Diagnosis: What the Research Shows

AI diagnostic support tools claim to improve differential diagnosis in primary care. The research is mixed: in highly controlled conditions (specific symptom clusters, narrowly defined diseases), some AI tools outperform general practitioners. In undifferentiated primary care presentations—the actual environment—the published performance is modest and often not better than a well-structured clinical reasoning process. The most common primary care presentations (fatigue, chest pain, abdominal pain) are high-dimensional, context-dependent problems where current AI performs below the clinical expectation.

AI-Generated Patient Communication at Scale

AI that generates patient-facing communications (after-visit summaries, discharge instructions, care plan explanations) shows promise but requires careful physician review. A 2023 JAMA study found that AI-generated patient instructions were accurate in 78% of cases—meaning 22% contained clinically significant errors requiring correction. AI-generated patient communication without mandatory physician review is a patient safety risk in its current state.

Predictive Analytics for Population Health

AI models that predict which patients will be hospitalized, develop complications, or disengage from care have been deployed in health systems since 2018. The models work at population level (they correctly identify higher-risk populations) but individual-level predictions remain imprecise. A model that identifies the 10% of patients with the highest hospitalization risk is useful for panel-level intervention prioritization; it should not be used to make individual care decisions without clinical assessment.

Evaluating AI Claims from EMR Vendors

AI Claim Evaluation Framework

Claim Type	What to Ask	Red Flags
"Reduces documentation time by X%"	What was the baseline? Measured how (time-motion study, self-report)? In what setting?	Only white paper evidence; no peer-reviewed study; dramatic percentages without methodology
"Improves clinical decision quality"	Compared to what? In what population? What is the false positive rate?	Claims based on narrow disease populations; no comparison to standard clinical practice
"AI meets HIPAA requirements"	Where is audio/data processed? Who has access? What is the retention policy? Is a BAA available?	Vague answers about data residency; BAA not provided upon request
"Reduces burnout"	What validated instrument? Over what time period? What is the control?	Testimonials instead of validated measures; short follow-up periods

How Krasyn Approaches AI in Primary Care

Krasyn builds AI that has a clearly defined role: assist physicians, not replace clinical judgment. The AI ambient scribe drafts notes; physicians review and sign. The billing review AI suggests codes; physicians confirm. The care gap dashboard surfaces open gaps; physicians decide how to address them. No Krasyn AI communicates with patients autonomously or makes clinical decisions without physician review.

We also publish what our AI cannot do accurately yet—the What Krasyn Does Not Do page is not marketing copy; it is a live list of documented limitations. See a demo of the AI clinical workflow.