ANNUAL INDUSTRY CONFERENCE
October 15–17, 2026 · Atlanta
See the conference · Get tickets

FOR SPONSORS • EXHIBITORS
October 15–17, 2026 · Atlanta
Request Prospectus • Supporters • Apply

PRACTICE DECISION TOOLS
FOR Doctors, Specialists, NPs, PAs
Calculators • Marketing • CMT Patient Study • All tools

RESEARCH & PROOF
National Stats, Industry Research
Research Briefs • Visuals • Legal Op • Natl Stats

AI in the Membership Practice: What the Trials Actually Show | CMT Research Brief No. 24

Educational content only. Not medical, legal, tax, financial or accounting advice. Read the disclaimer.

Concierge Medicine Today

CMT Research Brief No. 24 · Technology · October 2026

Research Brief No. 24TechnologyPrimary evidence: Clinical Trial Evidence

AI in the Membership Practice: What the Trials Actually Show

Ambient scribes, AI-drafted inbox replies and chatbots have now been tested in randomized trials. The results are real, modest, and more useful to a small-panel practice than the marketing suggests, if you know where to look.

By Concierge Medicine Today, Editorial ResearchPublished October 2026Version 1.1External review: pendingReading time: 12 minutes

Updated October 2026: evidence grades replaced with evidence types; funding disclosed on every source.

Start here: the questions this brief answers

Tap a question for the short answer, then jump to the evidence.

The 30-second answer

AI scribes have now been tested in randomized trials. They modestly reduce burnout and save minutes, not hours, per day. AI-drafted inbox replies tend to make messages better, not faster. For a membership practice, the case for AI is less about throughput and more about attention: giving the physician back eye contact and evenings. Adopt with a written policy, physician review of every note, and clear disclosure to members.

From the Editor-in-Chief

“AI scribes aren't the point. Attention is. If a tool gives you back time and you spend it looking at your patient instead of your screen, it's worth every dollar. If it just helps you see more people faster, you've quietly rebuilt the system you left.”

Michael Tetreault Editor-in-Chief, Concierge Medicine Today

Do AI scribes actually reduce physician burnout?

Yes, modestly. Several randomized trials published in 2025 and 2026 found lower burnout and work exhaustion scores. None found a dramatic change, and professional fulfillment did not reliably improve.

Go to the full answer ↓

How much time will an AI scribe save me?

The largest study so far, covering 8,581 clinicians, found about 16 minutes less documentation per 8 hours of patient care. Heavy users saved more, roughly 27 minutes.

Go to the full answer ↓

Will AI-drafted replies clear my inbox faster?

Probably not. Two studies found no reduction in reply time, and one found physicians spent more time reading. Replies did get longer and, by some measures, more empathetic.

Go to the full answer ↓

Does a small-panel practice benefit as much as a health system?

Nobody has studied concierge practices directly. CMT's reading is that the value shifts from time saved to attention restored, which matters more when attention is what members pay for.

Go to the full answer ↓

Do I have to tell patients I use AI?

In some states, yes. California requires a disclaimer on AI-generated clinical messages unless a licensed clinician reviewed them. Patients expect it either way: 72% in a 2026 Pew survey said being told is very important.

Go to the full answer ↓

What should I put in place before I buy a tool?

A one-page AI policy, a business associate agreement with the vendor, a rule that a clinician reviews every note, and a way to measure whether it helped. The checklist in Part 5 walks through it.

Go to the full answer ↓

Written for:Practice ownersPhysicians evaluating AI toolsPractice administratorsHealth system concierge leaders

Part 1 · The scribe evidence

The randomized trials are in, and they are modest

For two years, ambient AI scribes were sold on promise. In 2025 and 2026 the first randomized trials arrived. They agree on the direction and disagree on the size.

AI use is no longer an early-adopter story. In the AMA's 2026 survey of nearly 1,700 physicians, 81% said they use AI professionally, more than double the 2023 rate. Documentation tasks lead the list: 30% use it for care plans and progress notes, and 28% for charts and visit notes.¹

81%

of physicians use AI professionally, about double the 2023 rate

Practice Insight ¹

16 min

less documentation per 8 hours of patient care in the largest study (8,581 clinicians)

Population Data ²

51.9% → 38.8%

burnout before and 30 days after scribe use, six health systems

Practice Insight ³

9.5%

less time in notes with one scribe vs. control; the other showed no significant change

Clinical Trial Evidence ⁴

What the trials found

Study	Design	What changed	Evidence type
UCLA, NEJM AI 2025⁴	Randomized, 3 arms, 238 physicians, 14 specialties	One scribe cut note time 9.5%; the other did not. Both improved burnout, task load and work exhaustion modestly. Occasional clinically significant inaccuracies; one mild adverse event.	Clinical Trial Evidence
UW Health, NEJM AI 2026^5,6	Stepped-wedge randomized, 66 clinicians, 24 weeks	Work exhaustion fell; professional fulfillment did not significantly improve. The system expanded from 66 to more than 600 licenses.	Clinical Trial Evidence
Academic center, JAMIA 2026⁷	Randomized crossover, 160 clinicians, 2 products	Both tools reduced personal and work burnout. No meaningful change in after-hours "pajama time" or patient-related burnout.	Clinical Trial Evidence
Five academic systems, JAMA 2026²	Cohort, difference-in-differences, 8,581 clinicians	16.0 fewer documentation minutes and 13.4 fewer EHR minutes per 8 hours. Heavy users saved about 27 minutes. Quality and safety not measured.	Population Data
Six systems, JAMA Netw Open 2025³	Pre/post survey, 263 physicians	Burnout fell from 51.9% to 38.8% after 30 days.	Practice Insight

Minutes saved per 8 hours of patient care

Large five-system cohort study, ambient AI scribe adopters vs. non-adopters²

Analysis and chart: Concierge Medicine TodaySource: multisite JAMA cohort (2026), as summarized by ReachMD and Textbook of Digital Health. Observational design; documentation quality and patient outcomes were not assessed.

CMT reading

Every trial points the same way: less exhaustion, small time savings, and errors rare enough to manage but common enough to require physician review of every note. "Modest" is the honest headline. That is not a criticism. In medicine, a modest, reproducible benefit is often worth more than a dramatic, unreplicated one.

Part 2 · The inbox

AI-drafted replies make messages better, not faster

For a membership practice, the inbox is the product. Members pay for access, and access increasingly arrives as a portal message or a text. So the inbox evidence matters more here than almost anywhere.

Two early studies tested AI-drafted replies inside the electronic health record. At Stanford, clinicians used the drafts about 20% of the time. Reply time did not change, but task load scores fell from 61.3 to 47.3 and work exhaustion scores also fell.⁸ At UC San Diego, AI drafts did not reduce reply time; physicians spent more time reading, and replies were significantly longer.⁹

About one in five physicians (19%) now use AI to draft portal responses.¹

What it does well

Reduces the blank-page burden of starting a reply
Produces longer, warmer messages that some physicians then edit
Lowers self-reported task load

What it does not do (yet)

Cut time spent per message
Remove the need to read and verify each draft
Know your member the way you do

Membership-practice angle

A longer, more empathetic reply is arguably a feature in a model sold on relationship. The risk runs the other way: a polished AI reply can sound like the practice while saying something the physician would not. Review before send is not optional.

Part 3 · First principles

Why AI means something different in a small-panel practice

Every published trial was run in academic or system settings with full schedules. Concierge practices run on different math. CMT's analysis below is inference, not measured fact.

Start from the fundamentals. A high-volume physician's scarcest resource is time. A membership physician has bought back much of that time with a smaller panel. What members are paying for is attention: undistracted eye contact, a physician who remembers, a reply that sounds like a person.

Fundamental	High-volume practice	Membership practice (CMT inference)
What AI saves	Minutes that become more visits	Minutes that become presence in the room and evenings at home
How ROI is measured	Visits per day, revenue per hour	Retention, renewal rate, physician sustainability
Main risk	Errors at scale	Errors that damage a high-trust relationship; members noticing "the robot"
Where AI fits best	Notes, coding, inbox triage	Notes, pre-visit summaries, patient education drafts reviewed by the physician

In a membership practice, AI is not a productivity tool. It is an attention tool. Measure it that way.

That reframing changes the buying decision. A scribe that saves only 16 minutes a day may still be worth it if it lets the physician stop typing during a 45-minute visit. Conversely, a chatbot that answers members faster may cost more in trust than it saves in staff time.

What independent surveys show about adoption, cost and time

No published survey isolates concierge or membership practices, so the closest evidence comes from medical group surveys and health system reports.

83%

of medical groups use AI in patient visits, up from 71% a year earlier (189 practice leaders)

Practice Insight ¹⁰

46%

of practice leaders say AI tools improved provider productivity over two years; 27% saw no gain

Practice Insight ¹¹

$40 to $500+

per clinician per month: budget apps start near $40; enterprise platforms run several hundred dollars

Practice Insight ¹²

~$500

per provider per month quoted to one three-physician rural practice for EHR-integrated tools; it chose to wait

Practice Insight ¹²

Ambient documentation is the leading use case: in a 2025 MGMA poll of 351 practices, 68% added or expanded AI that year, led by ambient listening and note-taking, then scheduling and patient communication. Among practices not expanding, the common barriers were cost, unclear return on investment and EHR compatibility.¹³ The Peterson Health Technology Institute found strong signals on burnout but limited evidence of financial return, with most health systems reporting no increase in patient volume.¹⁴ Population Data

What this means for a membership practice

A small practice may pay listed prices rather than negotiated enterprise rates, and a membership practice rarely adds visits to recover the cost. At $100 to $500 per clinician per month, an AI scribe is a $1,200 to $6,000 annual expense per clinician. The question to ask is not “will it pay for itself in visits?” but “is the attention and time it returns worth that to our members and our physicians?” That framing is CMT's inference from the evidence above.

Part 4 · Trust and the law

Disclosure, consent and the patchwork of state rules

Physicians want guardrails as much as patients do. In the AMA survey, 86% called data privacy critical, 88% wanted robust safety validation, and 85% wanted a say in how AI is adopted. Clear liability rules ranked as the top regulatory priority.¹

State law is moving first. California's AB 3030, effective January 1, 2025, requires physician practices that use generative AI for patient communications about clinical information to include a disclaimer and instructions for reaching a human. The requirement does not apply when a licensed clinician reads and reviews the message. Physicians who violate it face licensing board discipline.¹⁵ Policy and Law

Patients expect to be told. In a June 2026 Pew Research Center survey of 3,488 adults, 72% said it is extremely or very important that providers tell them when AI is used in their care, rising to 81% for diagnoses. Only 16% knew a clinician had used AI in their care.¹⁶ Population Data In a UC Davis Health study of nearly 1,900 patients, 48% viewed AI scribes as a good solution, 33% were neutral and 19% had concerns; note accuracy was the top worry (39%), and 57% preferred to be told face to face.¹⁷ Practice Insight

Ambient scribes

Recording the visit. Most practices ask for verbal or written consent. State recording laws differ, and some require all-party consent. Ask counsel.

AI-drafted messages

Clinical content. Physician review before sending both protects patients and, in California, removes the disclaimer requirement.¹⁵

Patient-facing chatbots

Highest risk. The chatbot speaks for the practice without a clinician in the loop. Disclose clearly and route clinical questions to people.

Part 5 · Build

Before you buy: a seven-point AI readiness check

Use this as a conversation tool with your team and advisers. It is not legal advice and does not replace review by counsel.

Interactive checklist

How ready is your practice?

We have a one-page written AI policy that says which tools are approved and who may use them. Every vendor that touches patient information has signed a business associate agreement. A licensed clinician reviews every AI-generated note and clinical message before it is final. We have a consent script for recording visits that fits our state's recording laws. Our membership agreement or welcome packet tells members how we use AI. We measured a baseline (after-hours charting, burnout, member feedback) so we can tell if it helped. Staff know never to paste patient information into consumer AI tools.

Check the items you have in place.

“Healthcare may be a process, but the patient should never feel processed.” Michael Tetreault, Editor-in-Chief, Concierge Medicine Today

Source: CMT Media Desk.¹⁸ The test for any AI tool in a concierge practice: does it give the physician more attention for the patient, or less?

Learn

Read the trials, not the brochure

Expect modest time savings and real but small burnout relief. Ask vendors for peer-reviewed evidence, not testimonials.

Build

Pilot with a baseline

Run a 60 to 90 day pilot with two or three clinicians. Measure after-hours charting and how present you feel in visits, before and after.

Lead

Tell your members

Explain what AI does, what it never does, and that their physician reviews everything. Transparency is part of the membership promise.

Methods, limitations and evidence types

How this brief was built

CMT searched peer-reviewed journals, preprint servers, professional association surveys and state law summaries published through September 2026 for evidence on ambient AI scribes, AI-drafted patient messages and AI governance in ambulatory care. Priority went to randomized trials, then large multisite studies, then national surveys. Each figure was checked against the original publication or the best available summary of it; where CMT relied on a summary, the reference says so.

Adoption and cost figures come from MGMA practice-leader polls (self-selected respondents), trade press reporting of vendor price ranges, and a Peterson Health Technology Institute report. Patient disclosure figures come from a nationally representative Pew survey and a single-system UC Davis study. No study identified for this brief was conducted inside a concierge or membership-based practice. Conclusions about membership practices in Part 3 are CMT editorial inference and are labeled that way.

How to read the evidence types

Every key finding is labeled by evidence type. Labels describe the type of evidence, not its value. Each type answers different questions. Funding is disclosed on every source.

Clinical Trial Evidence
Randomized trials and systematic reviews.Best for cause and effect.

Population Data
Large observational studies and government data.Best for trends at scale.

Practice Insight
Surveys, smaller studies and expert consensus.Best for real-world experience.

Industry Research
Company-sponsored or company-reported data that is not peer-reviewed.Best for early signals and operating data.

Policy and Law
Statutes, regulation and official guidance.Best for what is required.

What we don't know

Whether AI scribes change outcomes patients care about, such as diagnostic accuracy, safety events or relationship quality. Most trials measured clinician experience and EHR time.
How AI tools perform in practices with 300 to 600 patients and 30 to 60 minute visits, where documentation patterns differ from high-volume clinics.
Long-term effects, including the skill-loss concern raised by 88% of physicians in the AMA survey.
Adoption, cost and member reactions in concierge and membership practices specifically; available surveys cover medical groups and health systems. Published price ranges come from trade reporting and change often.

How to cite this brief

Concierge Medicine Today. “AI in the Membership Practice: What the Trials Actually Show.” CMT Research Brief No. 24. October 2026. https://conciergemedicinetoday.net/ai-in-the-membership-practice

External review: this brief has not yet been reviewed by an outside expert. When review is complete, the reviewer is credited by name above with any conflicts of interest, and the version number is updated. Reviewers check accuracy and fairness; CMT is responsible for the final content.

Corrections policy: when an error is identified, CMT corrects it in the open and updates the version number above. Send corrections to the editor through conciergemedicinetoday.net.

Keep reading

Related CMT Research Briefs

Sources

References

American Medical Association. More than 80% of physicians use AI professionally: AMA survey (2026 Physician Survey on Augmented Intelligence). O'Reilly KB. AMA. March 12, 2026. www.ama-assn.orgFunding: not stated (conducted and published by the American Medical Association)
Multisite cohort study of ambient AI scribe adoption across five U.S. academic health systems, June 2023 to August 2025 (8,581 ambulatory clinicians). JAMA. 2026. Summary via ReachMD. reachmd.comFunding: not stated
Schwamm L, et al. Ambient AI scribes and clinician burnout across six U.S. health systems. JAMA Network Open. October 17, 2025. Summary via Yale School of Medicine. medicine.yale.eduFunding: not stated
Lukac PJ, Turner W, Vangala S, et al. Ambient AI scribes in clinical practice: a randomized trial. NEJM AI. 2025;2(12). Preprint and trial registration NCT06792890. www.medrxiv.orgFunding: not stated
Afshar M, et al. Pragmatic randomized trial of ambient artificial intelligence to improve health practitioner well-being (UW Health). NEJM AI. 2026. Summary via 2 Minute Medicine, February 2, 2026. www.2minutemedicine.comFunding: not stated
NIH Pragmatic Trials Collaboratory. Grand Rounds, January 30, 2026: A pragmatic randomized controlled trial of ambient artificial intelligence to improve health practitioner well-being (Afshar M, Baumann MR). rethinkingclinicaltrials.orgFunding: not stated
Comparing ambient scribes: a randomized crossover clinical trial addressing ambient scribe technologies' impact on physician burnout. Journal of the American Medical Informatics Association. 2026;33(5):990-999. pubmed.ncbi.nlm.nih.govFunding: not stated
Garcia P, et al. Artificial intelligence-generated draft replies to patient inbox messages. JAMA Network Open. March 20, 2024. Summary via HealthDay. www.healthday.comFunding: not stated
Tai-Seale M, Baxter SL, Vaida F, et al. AI-generated draft replies integrated into health records and physicians' electronic communication. JAMA Network Open. 2024;7(4):e246565. pubmed.ncbi.nlm.nih.govFunding: not stated
MGMA Stat poll. AI use expands; leaders shift attention to measurement (189 applicable responses). Medical Group Management Association. August 4, 2026. www.mgma.comFunding: not stated (conducted and published by MGMA)
MGMA Stat poll. Are AI tools making clinicians more productive? It's complicated (257 applicable responses). Medical Group Management Association. May 12, 2026. www.mgma.comFunding: not stated (conducted and published by MGMA)
Littrell A. Take note: the AI scribe era is here. Medical Economics. March 30, 2026. www.medicaleconomics.comFunding: not stated (news report)
MGMA Stat poll. Document, schedule, communicate: AI tools (351 applicable responses). Medical Group Management Association. September 30, 2025. www.mgma.comFunding: not stated (conducted and published by MGMA)
Peterson Health Technology Institute. Adoption of AI in Healthcare Delivery Systems: Early Applications and Impacts. March 2025. Summary via Fierce Healthcare, March 25, 2025. www.fiercehealthcare.comFunding: not stated (published by the Peterson Health Technology Institute)
ArentFox Schiff. California requires disclaimers for health care providers' AI-generated patient communications (AB 3030). 2024. www.afslaw.comFunding: not stated (law firm analysis)
Pew Research Center survey of 3,488 U.S. adults, June 22 to 28, 2026, on disclosure of AI use in health care. Reported by Becker's Physician Leadership, August 2026. www.beckersphysicianleadership.comFunding: not stated (Pew Research Center survey, reported by Becker's)
Leiserowitz G, et al. Patient perspectives on ambient AI scribes (nearly 1,900 patients, UC Davis Health). Journal of Medical Internet Research. 2025. Reported by TechTarget, December 17, 2025. techtarget.comFunding: not stated
Concierge Medicine Today. Media Desk: quotes from the Editor-in-Chief, FAQs and data. Accessed October 2026. conciergemedicinetoday.netFunding: Concierge Medicine Today (self-published; no outside funder) CMT original

Disclaimer

Educational and informational only. This CMT Research Brief does not constitute medical, legal, tax, financial, accounting or other professional advice, and it does not create a professional relationship of any kind. Statements about laws, regulations, tax rules and payer policies are general, may not reflect the rules in your state, and can change after publication. Consult a qualified attorney, accountant, tax adviser, compliance professional or licensed clinician before acting on anything here.

Independence. Concierge Medicine Today is an independent publication. It does not accept payment for favorable coverage, and it does not favor one practice model over another. Company names and products are mentioned for context only and are not endorsements. Funding is disclosed for every source in the reference list.

Accuracy. CMT verifies figures against their original or best available sources at the time of publication. Where a figure is an estimate, an inference or a company-reported number, the brief says so. This content is not without possible error or omission.