AI scribes saved modest time in Q2 2026 as payers and lawmakers contested AI-driven coding and denials

Health systems, practices and payers spent the second quarter finding out what ambient documentation and prior-authorization automation return. A five-system JAMA study measured the time saved, UnitedHealthcare tied rising costs to provider billing and AI revenue-cycle tools, at least 10 states enacted AI-in-health laws and Congress challenged Medicare's AI prior-authorization model.

Editorial snapshot: 31 July 2026

The central finding

AI scribes save clinicians modest time, and the added billing revenue measured at five academic centers, at Medicare rates, fell short of their estimated price, so larger claimed returns under fee-for-service rest on higher coding levels that payers now contest and states have begun to regulate.

01Key findings

What changed in health care AI in the second quarter of 2026

Four developments define the quarter: documentation tools connected to billing, a smaller and more consolidated funding market, a five-system measurement of what scribes save, and new legal limits on automated denials.

Capability

Scribes and EHRs reached into coding and claims

Abridge announced decision support, coding and real-time claims for its scribe platform on June 11, Oracle said an AI version of its Cerner electronic health record (EHR) system is coming, and UnitedHealth's Optum runs claims in real time. The clinical conversation and the claim are becoming one stream of data.

Capital

Digital health funding fell to $3.2 billion

U.S. digital health venture funding was $3.2 billion in the second quarter, down from $4.2 billion in the first quarter and $3.4 billion a year earlier, by Rock Health's count. Acquisitions reached 71, the most in a quarter since the third quarter of 2021, and revenue-cycle management was the most active consolidation area in the first half.

Adoption and economics

Scribes saved 16 documentation minutes per eight hours

A JAMA study of 8,581 clinicians at five academic health systems found that the 1,809 clinicians given scribes spent 16.0 fewer minutes documenting per eight scheduled patient hours and added 0.49 visits a week. Their evaluation and management (E/M) revenue rose about $167 a month at Medicare rates, below the $300 to $500 monthly seat price a January viewpoint estimated.

Accountability

States and Congress moved on automated denials

At least 10 states enacted AI-in-health laws between April 6 and June 2. Georgia, Iowa and Colorado barred insurer denials based on AI alone, and Maryland required reporting of AI use in denials. Members of Congress challenged Medicare's AI prior-authorization model through a Government Accountability Office determination they requested, Congressional Review Act resolutions and a House funding rider.

Operating question

Can your organization show, for each claim or denial, where AI shaped the documentation, the code or the decision, and which clinician reviewed it?

02Capability

Ambient documentation is being connected to coding and claims

The quarter's main technical change was integration. Scribe vendors, EHR vendors and payer-owned software began treating the encounter transcript, the billing code and the claim as one stream of data. At the same time, frontier models became more expensive at the top of the market, and two were withdrawn from all customers.

Scribe vendor · Announced June 11

Abridge's platform reaches into coding and claims

Abridge announced decision support, nursing documentation, revenue-integrity work with the American Health Information Management Association and real-time claims, and said it is building a clinical-conversation model with NVIDIA on Nemotron. It reported more than 300 health systems live and more than 100 million conversations a year; availability of each feature was not stated.

EHR vendor · Announced June 10

Oracle is rebuilding Cerner around AI

Oracle said an AI version of its Cerner hospital and clinic system is coming and that AI will cut doctors' computer time and lower costs. It forecast double-digit Oracle Health growth in fiscal 2027.

Payer platform · Company-reported, April 21

Optum automates the claims providers automate

UnitedHealth reported about 0.5 billion Optum Real real-time claim transactions year to date, a member chatbot called Avery and a digital prior-authorization tool. Payers are automating the same exchange from the other side.

More than 2.5BReal-time transactions expected in 2026

96%Early approval rate, digital prior-authorization tool

EHR vendor · Background · March 2026

Epic's native AI sets the comparison

Epic reported in March that more than 85% of its customers use Epic AI, after launching Chart with Art in February. It plans for its Penny agent to complete coding sessions and appeals more autonomously.

Frontier models · All industries

Frontier prices rose and two models were disabled

GPT-5.5 doubled GPT-5.4's list price to $5 and $30 per million input and output tokens, and OpenRouter measured effective cost increases of 49% to 92% on its traffic. On June 12 a U.S. directive barring foreign-national access led Anthropic to disable Fable 5 and Mythos 5 for all customers, routing new sessions to Opus 4.8 or a user's default model.

Cost and consequence

A scribe seat was estimated in January at $300 to $500 per clinician per month. As the transcript feeds coding and claims, buyers judge the purchase more on revenue-cycle effects, and a vendor's dependence on a single frontier model becomes a continuity question for clinical operations.

03Capital

Digital health funding fell in Q2 as acquisitions reached their highest count since 2021

U.S. digital health venture funding declined from the first quarter and from a year earlier, in a quarter when CB Insights counted $149.5 billion in global AI equity funding. Typical deals grew, large rounds took a larger share and no digital health company went public. Rock Health stopped labeling startups as AI-enabled, so its figures do not separate AI from other digital health funding.

$3.2B

U.S. digital health venture funding, Q2 2026

Q1 2026: $4.2B. Q2 2025: $3.4B.

71

Digital health acquisitions, Q2 2026

The most in a quarter since Q3 2021.

First half of 2026

Larger rounds took more of the capital

Investors put $7.4 billion into 244 deals in the first half. Twenty rounds of $100 million or more, at 19 companies, took 45% of the capital, up from 42% in 2025 and 22% in 2024; just over 8% of deals took nearly half the money.

$14MMedian deal, the highest since 2022 (2025: $12M)

Care-delivery AI · April

Abridge raised $316 million at a $5.3 billion valuation

Fortune reported the April Series E extension, which brings Abridge to about $830 million raised. Eli Lilly's strategic investment, amount undisclosed, was announced June 11; Abridge's chief executive told Fortune that clinical trials were the first priority of that relationship.

Consolidation

Revenue cycle led acquisitions, and no company listed

Rock Health counted 115 digital health acquisitions in the first half and named revenue-cycle management the most active consolidation area. No digital health company went public in 2026 through June; Oura filed for an initial public offering in May.

Payer budgets · Company-reported

Payers' AI budgets exceed any single venture round

UnitedHealth plans nearly $1.5 billion of AI initiatives in 2026, about a third for software products and platforms, and expects a conservative 2-to-1 return over the next few years, with many tools paying back within 12 to 18 months. Elevance cited more than $1 billion in digital and AI-enabled capabilities.

What it signals

Capital is consolidating around platforms that span documentation, coding and claims, and the largest AI budgets in care delivery sit with payers rather than startups. Rock Health's authors say AI has widened buyers' options and raised their demand for measurable returns, which favors vendors that can show results on a buyer's own claims.

04Products and delivery

Scribe vendors, EHR vendors and payers now sell overlapping platforms

Three supply models now compete for the same workflow. Specialist scribe vendors are extending into coding and claims, EHR vendors are building scribing and revenue-cycle agents into the record, and payer-owned software is sold to the hospitals whose claims the same parent company pays. In traditional Medicare, the Centers for Medicare & Medicaid Services (CMS) added a fourth model: contracted AI vendors reviewing prior authorization.

Specialist vendors

From scribe to documentation-to-claim platform

Abridge's June platform combines decision support that carries continuing medical education credit, nursing documentation, planned hospital-room integrations with Artisight and hellocare.ai, real-time claims, revenue integrity and clinical-trial screening. Aetna's chief medical officer and Cigna's chief information officer appeared on stage at the launch.

EHR vendors · Epic, March; Oracle, June 10

Scribing and coding agents inside the record

Epic's AI runs inside the record its customers already use, with a roadmap toward more autonomous coding and appeals, and Oracle is rebuilding Cerner around AI. For health systems on either platform, the alternative to a specialist scribe is a native feature.

Payer-owned software

One parent company sells to both sides of the claim

Optum sells AI coding and revenue-cycle software to hospitals, including Optum Integrity One, which Becker's describes as UnitedHealth's best-selling AI product. UnitedHealthcare, the insurer, builds tools to flag outlier provider billing.

Traditional Medicare

CMS contracted AI vendors for prior authorization

Since Jan. 15, the WISeR model (Wasteful and Inappropriate Service Reduction) has used AI-assisted review by six technology vendors, among them Cohere Health and Humata Health, in Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington. Polsinelli says the vendors are paid on contingency tied to denials.

Selected launches and announcements, April to June 2026

Status as stated by the company or agency at the time; the FDA count comes from Innolitics' review of April authorizations.

Selected launches and announcements, April to June 2026
Item DateStatusWhat it adds
Abridge: NEJM and JAMA content Date April 15 Status Partnership announced; content due in coming months What it adds Journal content for Abridge's decision support
FDA AI-enabled device authorizations Date April Status 27 authorized; 56% radiology What it adds Includes sepsis-flagging and ECG-AI devices
Optum Real and digital prior authorization Date Reported April 21 Status Live; company-reported What it adds Real-time claims; 96% early approval rate
Oracle: AI version of Cerner Date June 10 Status Announced What it adds AI-rebuilt hospital and clinic system
Abridge clinician intelligence platform Date June 11 Status Announced; feature availability not stated What it adds Decision support, nursing, coding and claims
WISeR AI-assisted prior authorization Date Since Jan. 15 Status Active; two services deferred in April What it adds AI review in traditional Medicare

Buying decision

A scribe purchase is becoming a platform decision. The vendor that drafts the note may also suggest the code and submit the claim, so revenue-cycle, compliance and EHR teams belong in the selection alongside clinical leaders.

05Adoption and economics

Scribe access is wide but use is uneven, and payer automation figures are self-reported

The quarter's adoption figures come from academic medical centers, vendors and payers, and each measures something different. None of them covers independent practices, community hospitals or critical-access hospitals.

Reported AI use and automation, by segment

Each row uses its publisher's own definition and population, so rows are not comparable with one another. PHTI is the Peterson Health Technology Institute.

Reported AI use and automation, by segment
Segment and source MeasureReported valuePeriod
Clinicians at five academic systems (JAMA) Measure Given ambient scribe access Reported value 1,809 of 8,581 Opt-in at four of five sites Period June 2023 to August 2025
Scribe adopters at those systems (JAMA) Measure Used it for half or more of visits Reported value About 32% Usage unknown for 25% Period June 2023 to August 2025
Large health systems (JAMIA survey cited by PHTI) Measure Adopted AI for clinical documentation Reported value All respondents Period 2025 publication
Epic customers (vendor-reported) Measure Using Epic AI Reported value More than 85% Period March 2026
Abridge customers (vendor-reported) Measure Health systems live Reported value More than 300 Period June 2026
UnitedHealthcare (payer-reported) Measure Prior-authorization requests electronic; in real time Reported value Nearly 95%; about 50% Period Q1 2026
Aetna (payer-reported) Measure Eligible requests approved within 24 hours; in real time Reported value More than 95%; more than 80% Period Q1 2026
All payers (PHTI estimate) Measure Prior-authorization transactions automated Reported value 40% Period Year not stated

Large systems and academic centers

Broad access, uneven depth

In the JAMA study, 43% of adopters used the scribe for fewer than half of their visits and 32% for half or more, with usage unrecorded for the rest; savings rose with use. These systems now choose between EHR-native AI and platforms spanning decision support, coding and claims, and they can measure use from EHR audit data.

Community and safety-net systems

Academic results may not transfer

The JAMA authors caution that their results may not carry over to higher-volume nonacademic practices, and the nearest evidence for these systems is vendor-reported. When the payback runs through coding that payers contest, finance and compliance belong in the scribe decision.

Independent practices

Exposed to downcoding without the gains

PHTI warns that providers not using AI absorb across-the-board downcoding without the documentation gains. Many insurers already pay outlier Level 4 and 5 evaluation and management (E/M) visits one level lower, an average difference of $50 a visit, Advisory Board reports. Indiana bars AI-only downcoding by insurers from July 1.

Payers

Automation claims now come with reporting duties

Payers' automation rates are self-reported. Maryland began requiring quarterly reports of adverse decisions by service and AI use on June 1, and Washington's reporting of the share of denials aided by AI begins Oct. 1.

Economics

Both sides have a reason to automate the same exchange: PHTI estimates a prior-authorization submission cycle costs plans $40 to $50 and providers $20 to $30. On the provider side, a scribe's economics depend on how deeply clinicians use it, which a system's own EHR audit data can show.

06Outcomes

Measured scribe savings were modest, and revenue estimates ranged from $167 to $1,004 a month

The JAMA study published April 1 is the quarter's largest independent measurement of AI in care delivery. Its time savings were modest and concentrated among heavy users and primary care clinicians. The revenue estimates measure different things: total E/M billing at five sites, billing units at one site and coding levels at one vendor's customer, produced by parties with different interests.

Peer-reviewed · JAMA, April 1

Five academic systems, 1,809 scribe adopters

Researchers compared clinicians using Ambience, DAX Copilot or Abridge at Mass General Brigham, Emory, UCSF, Yale New Haven and UC Davis with 6,772 nonadopters, using EHR audit data from June 2023 to August 2025. After-hours EHR time did not change significantly.

−16.0 minDocumentation time per eight scheduled hours

−13.4 minTotal EHR time per eight scheduled hours

+0.49Visits a week

Same study · Subgroups

Heavy users and primary care saved more

About 32% of adopters used the scribe for half or more of their visits; they saved 21.3 minutes of total EHR time per eight scheduled hours. Primary care clinicians saved 25.0 minutes of EHR time.

−27.3 minDocumentation, heavy users, per eight hours

−26.9 minDocumentation, primary care, per eight hours

Randomized trial · Background, November 2025

UCLA trial compared two scribes

In a randomized trial at UCLA Health, Nabla reduced time in notes by 9.5%, while DAX's 1.7% reduction was not statistically significant. Burnout and task-load improvements were secondary outcomes.

Customer-reported · Vendor release

Reid Health reported nursing gains with Abridge

Reid Health's chief nursing officer reported nursing vacancy falling from 18% to 8.6% and incidental overtime down 70% on rollout teams. The figures are a before-and-after account in Abridge's release, and the tool's contribution was not separated from other changes.

Payer-reported

Payers reported faster approvals and fewer denials

UnitedHealthcare said more than 90% of prior authorizations were approved within one business day on average, and Optum said its Optum Real platform can cut manual contact costs by 76%; Elevance reported denials down nearly 70%. An Optum Rx and Cleveland Clinic pilot reported 88% fewer appeals. The figures are company-reported and unaudited.

Three estimates of scribe revenue

Different measures, settings and payment rates; the estimates are not interchangeable.

Three estimates of scribe revenue
Measure Reported effectSettingOrigin
E/M revenue (visits and coding mix) Reported effect $167.37 per adopter a month 95% CI $86.52 to $248.21 Setting Five academic systems; 2025 Medicare rates Origin Peer-reviewed multisite study, JAMA; exploratory
Added relative value units (RVUs) Reported effect +5.8% weekly RVUs; about $3,000 a physician a year Setting One academic site; more than 1.2 million encounters Origin Peer-reviewed single-site study, January 2026 Dollar value from an accompanying commentary
Higher coding levels Reported effect $1,004 per provider a month Setting One Suki customer system; Level 5 +5%, Level 4 +7% Origin KLAS validation of Suki, cited by PHTI

For comparison, a January 2026 JAMA Health Forum viewpoint estimated a scribe seat at $300 to $500 per clinician per month.

Where the return comes from

The time savings support scribes as clinician relief. The two peer-reviewed revenue estimates fall below the $300 to $500 seat estimate; only the coding-level figure clears it, and it comes from one customer of one vendor and is the part of the return that payers contest.

07Corrections

Four scribe and prior-authorization figures that need context

Several figures circulated during the quarter in forms that change their meaning. Each one feeds purchasing cases, payer disputes or policy arguments.

Burnout · Out of date

The 40% figure is a March 2025 self-report from some health systems

Claim under review: AI scribes cut clinician burnout by up to 40%.

Becker's coverage of PHTI's March 2025 scribe report cited some health systems reporting burnout reductions of up to 40% among users. The April 2026 JAMA study measured modest time savings, and its senior author said those reductions are unlikely to fully explain changes in burnout; in UCLA's randomized trial, burnout was a secondary outcome.

Time saved · Overstated

Measured savings were 16 documentation minutes per eight scheduled hours

Claim under review: Ambient scribes save clinicians about an hour a day.

The five-system JAMA study found 16.0 fewer documentation minutes and 13.4 fewer total EHR minutes per eight scheduled hours, with no significant change in after-hours EHR time. Heavy users saved 27.3 documentation minutes.

Coding revenue · Source clarified

The $1,004 figure is a KLAS validation at one Suki customer

Claim under review: A multihospital system gained $1,004 per provider per month from scribe-supported coding.

PHTI's report cites the figure to KLAS' January 2026 validation of results at three customers of the scribe vendor Suki; the $1,004 came from one multihospital system where Level 5 encounters rose 5% and Level 4 encounters 7%. KLAS says its findings are not conclusive for Suki's whole client base.

WISeR · Initial rate

62% is the initial approval rate; about 84% were approved after physician review

Claim under review: WISeR approved 62% of prior-authorization requests in Texas.

Washington Post data cited by Georgetown's Center on Health Insurance Reforms show about 62% approved initially and about 84% after physician review. Comparators differ: Georgetown notes Medicare Advantage approval rates are often above 90%, and Polsinelli cites a 92% national average. CMS had not released comprehensive data.

08Accountability

Where prior-authorization automation and scribes fell short

Health care is following a pattern seen across industries: in a National Bureau of Economic Research survey of about 6,000 executives in four countries, fielded from November 2025 to January 2026, 89% reported no productivity effect from AI over the past three years. Care delivery differs in one respect: under fee-for-service, a tool that saves little time can still pay through coding, which turns an operational shortfall into a dispute with payers.

Incident · Traditional Medicare

WISeR turnaround and data gaps

Providers report spending up to 90 minutes to schedule a peer-to-peer review held a week or more later, and lawmakers say CMS' one-to-three-day turnaround covers only the vendor step. CMS deferred two services in April for operational readiness and had not released comprehensive data. The reviews' clinical accuracy has not been established either way.

Deployment depth

Savings were smaller where scribe use was light

After-hours EHR time did not fall significantly in the JAMA study, and only 32% of adopters were recorded using the scribe for half or more of their visits. Clinicians in UCLA's 2025 trial rated clinically significant inaccuracies in notes as occasional, which keeps clinician review in the workflow.

Risk · Prior authorization

Automation on both sides without system savings

PHTI found no evidence that AI lowers the cost per prior-authorization claim once the AI's own cost is counted, and warned of automated back-and-forth between payer and provider systems. AI may lower one organization's cost without reducing the system's, and real-time models are not yet scalable, PHTI found.

Risk · Coding intensity

Higher coding passes costs to patients and nonadopters

PHTI says coding intensity raises spending and patients' cost-sharing, and that payer downcoding may fall on providers who never adopted the tools. A March Blue Cross Blue Shield Association study, reported by Becker's, attributed about 20% of a 9% rise in commercial inpatient costs per member from 2023 to 2024 to coding intensity at hospitals with the sharpest growth in case complexity.

Risk · Agent security

Agent connectors widened the attack surface

The Cloud Security Alliance described a command-execution flaw in the Model Context Protocol (MCP) software development kit with an estimated 200,000 vulnerable instances in May, and in June an injection technique through untrusted error events that hijacked agents 85% of the time in tests. Agents connected to EHR or billing systems through such protocols share the exposure.

Who is accountable

State laws now assign responsibility to people: a clinician behind each AI-assisted adverse determination and, in Indiana, a human reviewer for each AI-generated claim. Payers and providers will need to show who reviewed what, decision by decision.

09Regulation

States required human review of AI-assisted denials while Congress challenged WISeR

Federal AI rulemaking for care delivery was light: by June 10, the FDA had authorized about 1,450 AI-enabled devices, apparently none of them using generative AI, according to the Congressional Research Service. The binding activity came from state legislatures and from congressional pressure on one CMS model. The EU and UK adjusted timelines and funding.

Insurer AI · States

A clinician behind AI-assisted denials

Washington, from June 11, and Arizona, from June 30, require licensed professionals to make medical-necessity denials. Iowa bars AI as the sole basis for denying, delaying or downgrading medical-necessity requests from July 1, Georgia and Colorado bar adverse determinations by AI alone from Jan. 1, 2027, and Alabama requires insurer AI to weigh each patient's history and circumstances, with annual certification, from Oct. 1.

Reporting · States

AI involvement becomes a reportable data element

Maryland requires quarterly reports of adverse decisions by service and AI use from June 1, and its insurance commissioner may investigate spikes, especially in emergency department denials. Washington's reporting of AI-aided denials starts Oct. 1, and Colorado will require disclosure and audits of utilization-review AI.

Claims and downcoding · Indiana

Limits on AI downcoding and review of AI-generated claims

Indiana's HB 1271, enacted in March and effective July 1, bars insurers from downcoding with AI alone and requires providers to have a person review AI-generated claims. It regulates both sides of the coding dispute in one law.

Federal · WISeR

Congress challenged Medicare's AI prior-authorization model

GAO determined May 12 that WISeR is a rule under the Administrative Procedure Act, opening Congressional Review Act procedures, and resolutions followed May 20. On June 9 the House Appropriations Committee adopted a rider barring funds for WISeR or any prior-authorization model in traditional Medicare. The rider is not law, and the model remained active at June 30.

Federal · Interoperability

Prior-authorization interfaces due Jan. 1, 2027

CMS-0057-F, finalized in January 2024, has required 72-hour and seven-day prior-authorization decisions since Jan. 1, 2026, and requires three new application programming interfaces by Jan. 1, 2027, with dates varying by payer type. PHTI wants EHR vendors required to integrate standardized prior-authorization interfaces and payers to publish machine-readable policies.

EU and UK

The EU moved to delay device AI duties, and the MHRA funded its sandbox

EU negotiators provisionally agreed May 7, and Parliament voted June 16, to move AI Act high-risk duties for products covered by Annex I, which includes the medical device and in vitro diagnostic regulations, from Aug. 2, 2027, to Aug. 2, 2028, and for Annex III systems from Aug. 2, 2026, to Dec. 2, 2027. Council adoption was pending at quarter end, so the original dates still stood. The Medicines and Healthcare products Regulatory Agency (MHRA) committed £3.6 million over three years to its AI Airlock sandbox.

State AI-in-health laws enacted or taking effect, April to June 2026

Enacted in the quarter unless marked earlier. Colorado repealed its general AI Act on May 14 before it took effect; HB26-1139 is a separate health law.

State AI-in-health laws enacted or taking effect, April to June 2026
State and law EnactedEffectiveMain rule
Oregon SB 1546 Enacted April 6 Effective Jan. 1, 2027 Main rule Chatbot disclosure and self-harm protocols; private suits
Tennessee SB 1580 Enacted April 6 Effective July 1, 2026 Main rule AI may not pose as a licensed mental health professional
Maine HB 2082 Enacted April 8 Effective Not stated Main rule Mental health AI for support tasks only; recording consent
Nebraska Conversational AI Safety Act Enacted April 8 Effective July 1, 2027 Main rule Chatbot disclosure and crisis protocols
Alabama SB 63 Enacted April 17 Effective Oct. 1, 2026 Main rule Insurer AI must use patient history; annual certification
Delaware HB 191 Enacted April 23 Effective Not stated Main rule AI may not use nurse, physician or PA titles
Maryland HB 1563 Enacted April 28 Effective June 1, 2026 Main rule Quarterly reporting of denials by AI use
Georgia SB 444 Enacted May 5 Effective Jan. 1, 2027 Main rule No AI adverse determination without peer review
Iowa HF 2635 Enacted May 13 Effective July 1, 2026 AI section Main rule AI not sole basis to deny, delay or downgrade
Colorado HB26-1139 Enacted June 2 Effective Jan. 1, 2027 Main rule No AI-only medical-necessity denial; audits, disclosure
Washington SB 5395 Enacted March 23 (earlier) Effective June 11, 2026 Main rule Only licensed professionals deny on medical necessity
Arizona HB 2175 Enacted 2025 (earlier) Effective June 30, 2026 Main rule Licensed professional decides medical-necessity denials

Compliance calendar

Dates already set at quarter end: Indiana, Iowa and Tennessee on July 1; Aug. 2 for the EU AI Act's Annex III duties unless the delay is adopted first; Alabama's insurer AI rules and Washington's denial reporting on Oct. 1; and Colorado, Georgia, Oregon, Utah, the rest of Washington's law and the CMS-0057-F interfaces on Jan. 1, 2027.

10Voices

Payers, clinicians, vendors and researchers on coding, burnout and denials

Three points draw broad agreement: scribes save some documentation time, more for heavy users and primary care; scribes and AI coding raise billing intensity; and prior authorization is being automated on both sides, with no one claiming systemwide savings. The voices divide on what the coding rise means. Nabla calls it accurate capture, UnitedHealthcare describes a billing pattern, and Daniel Polsky expects the gains to go to the most sophisticated providers rather than small or critical-access practices. They also differ on whether time savings explain burnout gains and on whether WISeR's turnaround figures tell the whole story. Advisory Board, which reported several of these remarks, is an Optum subsidiary.

Payers

Tim Noel

CEO, UnitedHealthcare

“This reflects continued service intensity and higher provider billing patterns”

In June he described AI revenue-cycle tools as a driver of cost trend that he does not expect to persist, Becker's reported. UnitedHealth's Optum also sells AI coding software to hospitals.

Payer · Q1 earnings call, April 21

Sarah London

CEO, Centene

London described the industry's joint prior-authorization effort as unprecedented and as self-disrupting rather than self-regulating.

Payers present cuts to prior authorization as voluntary reform. UnitedHealthcare pledged to cut medical prior authorizations by 30% or more, and Cigna removed hundreds of services from review and is exploring strategic alternatives for its eviCore unit.

Payer · earnings remarks reported by Becker's

Clinicians and health-system buyers

Dr. Rebecca Mishuris

Chief health information officer, Mass General Brigham; senior author of the JAMA study

“are unlikely to fully account for changes in burnout”

She was describing the study's time reductions. Her caution comes from a system that adopted scribes and helped produce the quarter's main evidence on them.

Health-system leader · study press release, April 1

Dr. Bobby DuPre

Chief medical information officer, FMOL Health ambulatory groups

Scribes raise costs three ways: clinicians see more patients, fuller notes support Level 4 or 5 billing where burned-out clinicians documented only to Level 3, and scribes suggest codes.

A buyer describes the coding mechanism payers contest partly as a correction of under-documentation. FMOL Health was one of the three Suki customers in KLAS' January 2026 validation.

Health-system buyer · reported by Advisory Board

Mac Boyter

Senior research director, KLAS

A chief medical information officer once enthusiastic about scribes as clinician relief now pushes for coding uplift, and CIOs, CMIOs and CFOs want financial returns.

Boyter co-authored KLAS' January 2026 validation of Suki's results, the source of the $1,004 figure.

Market researcher · reported by Advisory Board

Lawmakers and regulators

Rep. Suzan DelBene and 30 other members of Congress

Members of Congress

“CMS has not yet released comprehensive data on the WISeR model's implementation”

The letter asked for state-level turnaround, denial, appeal and peer-to-peer data by July 15. DelBene, Sen. Ron Wyden and Rep. Greg Landsman also led the Congressional Review Act resolutions.

Legislators · letter to CMS, June 22

Centers for Medicare & Medicaid Services

Federal agency running WISeR

CMS says a clinician reviews any AI-based denial before it is issued and that WISeR vendors are penalized for inappropriate denials.

CMS cites these safeguards and a fast vendor turnaround; lawmakers say the turnaround measure covers only part of the process.

Agency position · reported by Georgetown's Center on Health Insurance Reforms and Nelson Mullins

Vendors

Dr. Shiv Rao

CEO, Abridge

“We started Abridge to save time, save money, and save lives.”

Abridge's June platform adds coding, revenue integrity and real-time claims, which places the company on the billing side of the dispute as well as the documentation side.

Vendor · keynote release, June 11

Nabla

Ambient documentation vendor

“It is more accurate coding enabled by better documentation.”

Nabla cites a single-site peer-reviewed study that found no rise in claim denials, and an eight-week pilot at University of Toledo Health in which established Level 4 visits rose from 41% to 45% and Level 5 visits from 3% to 5%.

Vendor · company blog, May 27

Researchers and analysts

Dr. Lisa Rotenstein

UCSF; lead author of the JAMA study

The more clinicians use a scribe, the greater the benefit, a finding she offered health system leaders as a point of reflection.

Her emphasis on depth of use connects the study's modest average to the larger savings among heavy users.

Researcher · interview with TechTarget

Caroline Pearson

Executive director, Peterson Health Technology Institute

Investors, plans and providers privately agree that scribes raise coding intensity; insurers can pay for it, downgrade visits or cut rates.

PHTI's April report recommends disclosure of AI use in coding, audits, near-term price adjustments for Level 4 and 5 visits and diagnosis-related groups (DRGs) with sudden intensity rises, and later revaluation of CPT, RVU and DRG payments.

Independent institute · reported by Advisory Board

Daniel Polsky

Johns Hopkins University

The winners will be the most sophisticated providers; the tools are being adopted by the haves, not solo primary care or critical-access providers.

His view ties the coding dispute to the segments with the least measured evidence.

Health economist · reported by Advisory Board

Hannah Neprash

University of Minnesota School of Public Health

“scribes may also drive up health care costs for patients and the federal government.”

Her Medicare claims study of scribe spending is scheduled to finish in December 2026. With Paige Nong, she argued in January that billing-focused scribes may raise spending without benefit to patients.

Health economist · university release, February (background)

11Decisions

Three decisions for finance, clinical informatics and utilization management

Chief financial officer

Whether to count coding uplift in the scribe business case

The added E/M revenue measured at five academic centers was about $167 a month at Medicare rates, against a $300 to $500 seat estimate. The coding-level case rests on one customer in a KLAS validation of Suki, and payers already answer outlier Level 4 and 5 claims by paying one level lower.

Would the scribe contract still pay back if outlier visits were paid one level lower?

Chief medical information officer

How to raise and measure the depth of scribe use

Savings concentrated among the roughly one-third of adopters who used scribes for half or more of their visits, and among primary care clinicians. After-hours EHR time did not change.

Which clinicians would gain most from deeper use, and does the audit data exist to show it?

Payer chief medical officer

How AI-assisted denials are reviewed and recorded

Washington, Arizona, Georgia and Colorado require clinician or licensed-professional review of AI-assisted denials, and Maryland and Washington require reporting of AI involvement. Congress is scrutinizing the same practice in WISeR.

For every adverse determination, can you show whether AI was involved and which clinician decided?

12Actions

Four actions for the next quarter

  1. 1

    Build a scribe-use baseline from EHR audit data

    Track documentation time, after-hours time and the share of visits using the scribe by specialty, as the JAMA team did with Epic Signal data. The baseline shows where adoption gaps sit and which clinicians benefit.

    Clinical informatics

  2. 2

    Monitor coding-level shifts after rollout

    Compare Level 4 and 5 distributions, denials and payer edits before and after scribe deployment, and record human review of AI-suggested codes. The record supports payer audits and rules like Indiana's on AI-generated claims.

    Revenue cycle and compliance

  3. 3

    Log AI involvement in every determination

    Record whether AI was involved and which clinician decided as one data element across utilization-management, claims and EHR systems. It serves Maryland's reporting now, Washington's from Oct. 1, and Colorado's and Georgia's rules from Jan. 1, 2027.

    Payer utilization management

  4. 4

    Map vendor model dependencies and agent connectors

    Ask clinical and revenue-cycle vendors which frontier models they use and how they fall back, as Greenberg Traurig advised after the June 12 directive, and review agent connectors for the MCP flaws reported in May and June. The result is a known supply and security exposure before prior-authorization interface work for 2027.

    Chief information officer

13References

References

This briefing covers April 1 to June 30, 2026, including reports published by July 31, 2026, that describe that period. Earlier studies, rules and laws are cited as background where they explain a change in the quarter.

Payers and providers

Federal policy and WISeR

EU and UK

Thoughtful engineering for what’s next.

Have a question about what this means for your team? Let’s talk it through.

Start a conversation

Have a project in mind, or just exploring an idea? Let’s talk.
Fill out the form or give us a call.

413-570-3299

What can we help you with?