Can AI Save the Primary Care Exam Room From Quality Creep?

Years ago, a practice operations leader told me that every patient encounter — no matter how urgent — was an opportunity to capture a quality metric. She referred to this as “in-reach,” complimenting outreach to patients when they are not in the office.

“Even when someone is acutely ill?” I asked. “Even then,” she said.

The implication was clear: in the middle of treating a patient with gastroenteritis, I should also be thinking about overdue mammograms.

At the time, I was appalled. Today, as a medical director for a large primary care network, I better understand the forces behind that mindset. I help lead efforts to meet system-wide quality targets tied to value-based payment and reputation. I see their importance. And yet, the discomfort remains. The exam room, whether physical or virtual, is an intimate space where clinicians should be fully focused on the patient in front of them. Increasingly, it is also a place where competing agendas intrude.

As generative artificial intelligence (AI) rapidly enters primary care workflows, a new question is emerging: Will these tools restore the doctor-patient relationship or further entrench the metrics that have strained it?

To answer that, it’s worth revisiting what problem quality measurement was originally meant to solve. Early efforts were designed to benchmark best practices, standardize care, and improve outcomes while lowering costs. Over time, however, quality measurement has become tightly coupled to reimbursement. What began as a clinical improvement tool has evolved into a documentation exercise that often prioritizes what is measurable and billable over what matters most to patients.

The consequences are not theoretical. Family physicians in a recent qualitative study by Richard Young, MD, et al. described moral distress when quality metrics interfered with their ability to provide patient-centered care. Many clinicians experience this tension daily, balancing the needs of the individual patient with the demands of population health targets embedded in electronic workflows.

Now AI promises relief. Today’s tools can summarize charts, draft messages, surface care gaps, and automate documentation. A growing industry of companies is going further, using AI to identify data that satisfy payer metrics and generate “audit-ready” notes. For overburdened clinicians, this can feel like a lifeline.

But these tools do not solve the underlying problem — they industrialize it.

Automating compliance with flawed metrics risks scaling the very pressures that erode the exam room. The “wow” factor of AI-generated documentation may distract from a more fundamental question: Are we even measuring the right things? Polishing the ruler does not fix a broken measurement system.

There is also a competitive dimension that makes this harder to ignore. Health systems operate on thin margins, and value-based payment programs reward those who document metrics most effectively. AI tools that accelerate metric capture risk becoming the anabolic steroids of the quality game. If everyone else is using them, those who abstain may fall behind — not because they deliver worse care, but because they document it less aggressively.

This dynamic should give us pause. When success depends on optimizing documentation rather than improving care, we risk widening the gap between what patients need and what systems reward.

None of this is an argument against measurement itself. Accountability matters. But the path forward requires re-centering quality measurement on outcomes that patients value and clinicians can meaningfully influence. That means emphasizing results over process, crediting shared decision-making, and recognizing that informed patients may reasonably decline recommended screenings or treatments.

It also means acknowledging the realities of primary care. Many practices operate with limited staff, constrained access, and significant administrative burden. Quality reporting should be straightforward, with data easily extracted from the electronic health record, not dependent on layers of manual work that pull clinicians and staff away from patient care.

So where does AI fit in a better version of this system?

Used thoughtfully, it can help. AI-assisted chart review can improve risk adjustment and highlight social determinants of health that meaningfully affect outcomes. Pre-visit planning tools can identify care gaps before the patient arrives, allowing teams to address them efficiently when appropriate. Automated outreach, whether through portals, text messaging, or other channels, can shift population health work outside the visit, preserving the encounter for the patient’s immediate concerns.

Just as important, technology should reduce — not intensify — the surveillance of clinicians. Excessive scoring and monitoring of individual performance against system-wide metrics can drive burnout, encourage gaming, and deepen the divide between clinical and business priorities. AI should aim to lower the background noise, not amplify it.

Looking back on that early conversation about “in-reach,” I still believe there is a role for it — but not as a checklist imposed on every encounter. Instead, in-reach should be patient-driven, grounded in education and shared goal-setting. When preventive care is discussed, it should emerge naturally from the patient’s context, not because a metric demands it in that moment.

In the throes of acute illness, a patient does not need a reminder about a screening test. They need attention, empathy, and care. No algorithm should compete with that. And no quality system should demand it.