AI scribes in medical clinics: six myths, and what the evidence actually shows

Ask almost any family doctor what wears them down and documentation will come up quickly. Notes, referrals and inbox messages follow clinicians home and eat into time with patients. That is why ambient AI scribes have spread so fast. An ambient AI scribe is software that listens to a visit (with the patient’s permission), turns the conversation into text, and drafts a structured clinical note for the clinician to review and sign.

Adoption is broad. In the American Medical Association’s 2026 survey of 1,692 US physicians, 72% said they had incorporated at least one AI use case into their work, up from 38% in 2023, and 28% used AI for billing codes, charts or visit notes ¹⁵. In Canada, Canada Health Infoway’s national program enrolled 12,019 primary care clinicians between June 2025 and February 2026, and monthly AI-assisted encounters grew from over 244,000 to over 1.22 million ⁵.

Clinic operations leaders are now hearing big promises from vendors and mixed reports from colleagues. Below are six common beliefs, set against the best evidence available in 2026.

Myth 1: "An AI scribe gives every doctor back hours a day"

Reality: the measured time savings are real but modest, usually minutes per day or about an hour or two per week.

The strongest evidence comes from randomized trials, where clinicians are assigned by chance to use a tool or not. At UCLA Health, 238 physicians across 14 specialties were randomized to Microsoft DAX, Nabla or usual care. Nabla cut average time per note from 4 minutes 30 seconds to 3 minutes 49 seconds, a reduction 9.5% larger than in the control group. DAX’s smaller decrease was not statistically significant ¹. At Duke, a crossover trial in which 136 clinicians tried two anonymized products found that the better one reduced time in notes by 3.19 more minutes per day than the other ².

Large real-world programs point the same way. Canada’s national evaluation found clinicians saved 1.11 hours per week during work hours and 0.87 hours after hours at six months ⁶. Kaiser Permanente’s Northern California rollout, covering 7,260 physicians and about 2.58 million encounters, saved nearly 16,000 hours of documentation time over 15 months ⁴. That is a large total, but spread across so many visits it works out to well under a minute per encounter.

For planning, assume savings of minutes, not hours, and expect them to vary widely by clinician. Kaiser found its top third of users accounted for 89% of scribe activations ⁴.

Myth 2: "If it doesn’t save hours, it isn’t worth it"

Reality: the clearest benefit so far is lower cognitive load and burnout, not raw time.

In a study across six US health systems, the share of clinicians reporting burnout fell from 51.9% to 38.8% after 30 days of using an ambient scribe ³. That study had no control group and was supported by the vendor, so it likely overstates the effect. The randomized UCLA trial, which did have a control group, found a more modest improvement of about 7% in burnout scores ¹. In Duke’s trial, both tools reduced burnout scores ².

Canadian clinicians reported similar experiences: almost 80% said cognitive load was lower and nearly 70% said administrative burden was lower. Same-day note completion rose from 55.9% to nearly 75% ⁵. For a clinic struggling to recruit and keep physicians, reduced mental strain may matter more than a few minutes on the clock. It is also harder to measure, so plan to track it deliberately.

Myth 3: "The notes are accurate enough to sign without reading"

Reality: AI notes are drafts. They contain errors, and the hardest errors to catch are omissions.

A Veterans Health Administration study published in the Annals of Internal Medicine in April 2026 compared notes from 11 AI scribe tools with notes written by 18 human clinicians for five standardized primary care visits. Thirty blinded raters scored human notes higher in every case ⁷. The authors concluded that AI scribes should be treated as tools for generating draft documentation that needs review and editing.

In the UCLA trial, clinicians reported that notes "occasionally" contained clinically significant inaccuracies, most often omissions or pronoun errors, and one mild patient safety event was reported ¹. Omissions are especially risky because a note does not show what is missing ⁹. In a simulated study of medication histories, an audio-only scribe captured 81% of details correctly, compared with 98% when video was added ⁸. Simulations are not real clinics, but the gap shows how much a scribe can miss when it only hears the conversation.

Myth 4: "It will pay for itself through more visits"

Reality: so far, most clinicians use the time to catch up, not to see more patients.

The Peterson Health Technology Institute, an independent evaluator, reviewed early deployments in 2025 and concluded that AI scribes had not yet achieved significant return on investment. Provider organizations reported no increase in patients seen ¹³. In the six-system US study, clinicians’ willingness to see additional patients did not change ³.

Canada’s evaluation found the same pattern. Over half of the time clinicians saved went to other administrative tasks, about a fifth to personal time, and nearly 20% to seeing more patients ⁶. Vendor claims of extra visits and revenue are worth testing against your own baseline before they go into a business case.

Myth 5: "Higher billing levels are pure upside"

Reality: more complete notes can support higher coding, and payers are already pushing back.

Scribes capture more of what was discussed, which can support higher evaluation and management (E/M) billing levels in fee-for-service systems such as most US practices. Trilliant Health, a health analytics firm, examined claims at six health systems that announced ambient scribe adoption. Between 2018 and 2024, the share of new-patient visits billed at the two highest levels rose by 12 to 20 percentage points ¹⁰. Trilliant notes the analysis has no control group, that billing rules changed in 2021, and that the cause of the rise is not clearly understood.

Insurers are responding anyway. A policy brief in npj Digital Medicine describes the risk of a "coding arms race" and notes that Cigna began automatically reducing many high-level E/M claims by one level from October 2025 unless documentation supports them ¹¹. The Blue Cross Blue Shield Association linked a rise in certain maternity diagnoses, without matching treatment, to about $22 million in additional spending in 2023–2024 ¹². The lesson for US clinics is simple: the clinician signs the claim, so the clinician must check what the scribe added. Canadian clinics on fee-for-service provincial billing face a smaller but similar question about whether documentation supports the fee code.

Myth 6: "It’s just an IT install"

Reality: an AI scribe records patient conversations, so it needs privacy, consent and governance work up front.

In early 2026, Ontario’s Information and Privacy Commissioner issued guidance for health custodians on AI scribes. It calls for a privacy impact assessment before deployment, a governance body with authority to pause or modify systems, data minimization, vendor obligations to report unexpected system behaviour, staff training and human oversight ¹⁴. In the US, the AMA survey found 86% of physicians rated data privacy as important and 88% rated validation of safety and effectiveness as important ¹⁵.

Patients mostly accept the tools when asked. In the UCLA trial, fewer than 10% declined ¹. In a Kaiser Permanente patient survey, two-thirds were comfortable and 8% reported some discomfort ⁴. That acceptance depends on asking clearly and respecting the answer.

Where human oversight must stay

• Every note is reviewed before signing. Turn off any auto-accept feature. As UCLA’s senior author put it, the technology "requires active physician oversight, not passive acceptance" ¹.
• Diagnoses and billing codes get a separate check. Spot-audit a sample of notes against the visit to catch added or missing details ¹¹.
• Consent is explicit and documented, with an easy way for patients to say no.
• Errors are reported and acted on. Give staff a simple route to flag problems without blame ¹⁴.

Practical next steps for clinic leaders

1. Measure a baseline first. Record time in notes, after-hours charting, same-day note completion and a short burnout survey before go-live.
2. Run a time-limited pilot with two products if you can. The Duke and UCLA trials both found meaningful differences between tools ¹².
3. Complete a privacy impact assessment and confirm where audio and transcripts are stored, for how long, and whether the vendor uses them for training.
4. Write a one-page consent script for front-desk staff and clinicians.
5. Set review rules for notes, diagnoses and codes, and audit a small random sample each month.
6. Decide in advance how saved time will be used, whether that is shorter days, inbox work or added appointments, so the benefit does not quietly disappear.
7. Review results at 90 days against your baseline before expanding licences.

Sources

1. UCLA Health (November 26, 2025). UCLA study finds AI scribes may reduce documentation time and improve physician well-being. On Lukac et al., "Ambient AI Scribes in Clinical Practice – A Randomized Trial," NEJM AI (2025). Randomized trial, 238 physicians, 14 specialties.
2. Chowdhury et al., Journal of the American Medical Informatics Association (February 23, 2026). Comparing ambient scribes: a randomized crossover clinical trial addressing ambient scribe technologies’ impact on physician burnout. Duke. 160 clinicians enrolled, 136 analyzed.
3. Olson et al., JAMA Network Open (October 2, 2025). Use of Ambient AI Scribes to Reduce Administrative Burden and Professional Burnout. Pre/post quality improvement study, 263 clinicians, 6 US health systems. No control group. Vendor-supported data collection.
4. Kaiser Permanente Division of Research (April 1, 2025). AI-assisted notetaking gains steady support from Kaiser Permanente physicians. 7,260 physicians, 2,576,627 encounters, October 2023–December 2024. Surveys of 102 physicians and 118 patients.
5. Canada Health Infoway (June 23, 2026). Canada Health Infoway Releases Results of National AI Scribe Program. Program sponsor’s release. 12,019 clinicians enrolled.
6. Ha et al., Women’s College Hospital Institute and Canada Health Infoway (April 2026). Measuring the Impact of AI Scribes in Canadian Primary Care: Findings from a Nationwide Evaluation. Surveys of 3,678 (baseline) to 1,899 (six months) clinicians. Infoway-funded.
7. UW Medicine Department of Medicine (April 17, 2026). AI scribe tools produce lower quality medical notes compared to human clinicians. On Reddy et al., Annals of Internal Medicine (2026). 11 tools, 18 clinicians, 5 standardized visits, 30 blinded raters.
8. Menz et al., npj Digital Medicine (February 26, 2026). Vision-Enabled AI scribes reduce omissions in clinical conversations: evidence from simulated medication histories. Simulated interviews, 100 test recordings.
9. Sorin and Klang, JMIR Medical Informatics (August 10, 2026). AI Scribe Safety: Measuring What Happens After Signing. Letter. No new data.
10. Trilliant Health (March 12, 2026). Increased Outpatient Coding Intensity Following Hospital Adoption of AI-Enabled Scribing Warrants Examination. Analytics vendor. All-payer claims, six health systems. No control group.
11. Dai, Kvedar and Polsky, npj Digital Medicine (December 24, 2025). Policy brief: ambient AI scribes and the coding arms race. Policy comment. No original data.
12. ACDIS (April 23, 2026). AI scribes indicate a shift in coding intensity and reimbursement, new research finds. Trade news summarizing BCBSA and Trilliant reports.
13. Fierce Healthcare (March 2025). Early evaluation of AI scribes finds decreased burnout but limited financial ROI. On the Peterson Health Technology Institute report (8 health systems, 10 vendors).
14. Baker McKenzie (March 20, 2026). Canada: Ontario AI Scribe Guidance Signals Emerging Regulatory Baseline. Law firm summary of the Ontario IPC’s "AI Scribes: Key Considerations for the Health Sector."
15. American Medical Association (March 2026). 2026 Physician Survey on Augmented Intelligence. 1,692 US physicians, January 15–February 2, 2026. Self-reported.