Ambient AI scribes are boosting provider revenue. Insurers are already clawing it back.

The ambient AI scribes that hospitals adopted to ease clinician burnout have turned into a new leading driver of rising healthcare costs. By drafting fuller clinical notes, the tools let providers bill more for the same care, a shift health plans rank among their top cost pressures for 2027 in PwC’s latest Behind the Numbers report.

PwC expects commercial medical cost trend, the year-over-year change in what it costs to treat the same patient, to reach 9 percent for employer plans and 8.5 percent for individual coverage in 2027, its highest level in 17 years. The firm bases the estimate on actuaries at 27 health plans that together cover more than 103 million employer-sponsored members and 8 million Affordable Care Act enrollees. Nearly 70 percent of the actuaries placed AI documentation and coding tools among their top three cost drivers, and about one in five called the technology the biggest one.

When a note records more detail, a clinician can justify a higher-level evaluation and management code, the rubric that sets how much a visit pays. A visit documented at level 3 can move up to level 5, for example, which pays more for greater complexity, despite the actual care provided remaining the same. PwC’s actuaries describe documentation that captures greater specificity and reimbursable severity without a matching increase in the work delivered, a billing shift DHI’s Side Effects newsletter has examined.

AI scribes are now driving revenue

The bind for hospital technology leaders is that the scribes driving the trend are the products they championed for clinician well-being. A policy brief in npj Digital Medicine, written by researchers at Johns Hopkins and Harvard Medical School, traces how these tools moved from pilots into routine use at large systems even as their business case shifted toward revenue. Vendors still cite the independent evaluations behind these tools, which measure less documentation time and lower burnout, even as the pitch they build on that evidence moves from saving time to generating revenue.

The actual revenue effect of ambient AI scribes is easy to measure. A UCSF study in JAMA Network Open tracked 1.2 million ambulatory visits across 1,565 physicians at a single site and found that those using AI scribes billed 1.81 more relative value units per week, the unit Medicare uses to price physician work, than colleagues who did not. They also saw about 0.80 more patients per week, with no rise in denied claims. Through the 2025 Medicare fee schedule, that works out to roughly $3,044 per physician a year, modest against a doctor’s billings. The per-visit change is tiny, 0.04 units, but it holds across millions of encounters.

Higher AI coding isn’t always overbilling

The data appears to lend itself to a fraud narrative around management codes being raised deliberately with assistance from AI scribes, but the evidence does not support this framing. PwC ranks its inflators by how far each departs from the historical trend, not by total dollars. AI tops the list as the newest and fastest-moving pressure, but not the largest. By PwC’s own account, labor, supplies, and overall utilization still account for more of the 9 percent increase in cost. The UCSF authors are explicit that they cannot say whether the added units reflect more services, more accurate coding, or genuine overstatement, and they call for randomized work to determine the cause.

The npj brief makes the same point: when documented conditions rise, the cause is often details that clinicians previously omitted. At Virginia’s Riverside Health, clinicians using one AI scribe recorded 11 percent more work units and 14 percent more risk-adjustment diagnoses per visit, and a 2024 Texas Oncology analysis found documented diagnoses climbed from 3.0 to 4.1 per encounter. Determining whether these results capture real or inflated complexity requires a chart-by-chart analysis.

How insurers respond to AI coding

Insurers have already started clawing the money back. As the npj authors document, Cigna began automatically reducing many mid- and high-level evaluation and management claims by one level on October 1, 2025, unless the documentation clearly supports the higher code. Aetna Better Health applies similar reviews. The authors frame this back-and-forth as a coding arms race. Providers capture more, payers tighten audits and cut base rates at renewal, and the early advantage erodes as the practice spreads.

This dynamic rewards size, and large systems with documentation-integrity teams and capital can tune these tools and defend their coding. Meanwhile, small practices and safety-net clinics risk adopting AI scribes late, missing the temporary upside of higher level coding, and then operating under rates already reset downward by insurers. An AI tool originally meant to reduce the workload for clinicians suddenly becomes a financial variable that payers actively price against when contracts come up for renewal.