Are AI medical scribes worth it?
Scribeable, which makes an AI medical scribe, gives this answer: an AI scribe is worth it when you can check its note faster than you could have written your own, and the published studies show that this varies. In one study of 263 ambulatory clinicians across 6 US health systems, 30 days with an ambient scribe was associated with burnout falling from 51.9% to 38.8%. In a study of 902 outpatient clinicians at a Boston health system, enrollment was associated with 1.21 fewer minutes of note-active time per encounter, with no difference in after-hours note closure or work RVUs.
So expect a lighter cognitive load before you expect your evenings back, and test it on your own clinic days rather than taking any vendor's word for it, ours included. There is no published study of Scribeable, and we do not publish a time-saved number.
14-day Pro trial, no credit card required. Everything in Pro, including Rounding, for up to 150 notes. To keep Pro after day 14, subscribe for $79/month plus applicable tax. If you don't, nothing is charged and your account moves to the Free plan once you add your NPI (until then it is read-only), with every note kept.
We make Scribeable, so read this with that in mind. Every study below is linked so you can check it yourself. Three of them name products that compete with ours, and we report those results as published.
What the published studies found
| Study | What was studied | What it found |
|---|---|---|
| Olson et al. 2025, JAMA Network Open (PubMed 41037268) | Quality-improvement study, 263 ambulatory clinicians, 6 US health systems, 30 days with an ambient AI scribe. | Associated with burnout falling from 51.9% to 38.8%, a 2.64-point drop in note-related cognitive task load on a 10-point scale, 0.9 fewer hours of after-hours documentation, and more reported ability to add urgent patients to the schedule. |
| Booth et al. 2026, Annals of Internal Medicine (PubMed 42932880) | Retrospective interrupted time series of an ambient scribe (Heidi), 902 outpatient clinicians in a 14-hospital Boston health system, July 2024 to January 2026. | Associated with 1.21 fewer minutes of note-active time per encounter, 2.50 fewer among high users. High users' notes grew longer. No differences in same-day note closure, after-hours note closure, or work RVUs. |
| Lukac et al. 2025, NEJM AI (DOI 10.1056/AIoa2501000) | Head-to-head randomized trial at UCLA. Both arms reviewed and edited draft notes before signing. | Nabla was associated with a 9.5% lower documentation burden. DAX showed no significant change. |
| Haberle et al. 2024, JAMIA (PubMed 38345343) | Cohort study of Nuance DAX ambient documentation. | No quantifiable effect on patient safety and no productivity benefit. |
| Palm et al. 2025, Frontiers in Artificial Intelligence (PubMed 41199808) | Peer-reviewed audit of an LLM ambient scribe. | Reviewers detected hallucinated content in 20% of physician-written notes and 31% of AI-generated notes. |
These are studies of other products and of the category. Quality-improvement and cohort designs show association, not cause.
What those results mean for a practicing clinician
- The most consistent finding is less mental load. The largest time differences are small per encounter, and they do not always reach the evening.
- A scribe does not remove the review. In the UCLA trial both arms still reviewed and edited before signing, and that review is where a wrong line gets caught.
- Notes can grow. If a longer note is harder to read and to code from, a shorter visit in the room can turn into a longer chart.
- Humans also put wrong things in notes. The fair comparison in the Palm audit is 31% against 20%, not 31% against zero.
Where AI scribes go wrong, and what to look for
| Failure | What it looks like | What to look for in a product |
|---|---|---|
| A line nobody said | A dose or a history detail that was not in the visit. | A way to open the source of a sentence. In Scribeable, each sentence drawn from the visit or the chart links back to where it came from. |
| A missing detail filled with a guess | A plan with a dose or laterality the visit never stated. | A question instead of a guess. Scribeable asks a short question when the plan needs a detail nobody said. |
| A note that does not fit | The wrong sections, in the wrong order. | Your own format. Scribeable builds a template from a note you paste in. |
| A note the coder cannot use | Missing specificity, no support for the level. | Suggested codes with the matching note text, for you or your coder to check. |
A two-week test you can run yourself
Any scribe, including ours, should pass this on your own patients before you pay for it.
- For a normal week, write down when you close each note and how long your evening charting runs. That is your baseline.
- Run the scribe on the same kinds of visits for two weeks. Keep your old method available.
- For each note, count the lines you changed and whether any line was wrong rather than merely worded differently.
- Open the source of at least three sentences a day and check that they say what the note says.
- If you bill, have your coder read a sample and tell you whether the level is supported.
- At the end, compare time per note and evening charting to your baseline. Keep it only if both moved, or if the load feels lighter enough to pay for.
If you are charting every night
A scribe helps most when the bottleneck really is writing. Before you buy one, check where your evening goes. Inbox messages, refill requests and prior authorizations will not move because the note got easier.
- Close the note between patients while the visit is fresh, even if it is rough. A note written that evening is written from memory.
- Say the assessment and plan out loud in the room. A scribe can only write what it heard, and the patient hears the plan too.
- Use one template per visit type, in your own format, so you read in the same order every time.
- Move what is not documentation off your plate. On the Scribeable Practice plan, the front desk logs refill requests and callbacks into one queue, and the coder works from a queue of signed notes.
Is it worth it for your practice?
| Practice | What decides it | A place to start |
|---|---|---|
| Solo | Whether you can review faster than you write, and whether you want coding help. | Scribeable's Free plan writes notes in your templates at $0 with no monthly note cap once you add your NPI. Pro is $79 a month with codes, letters and calculators. |
| Small group | Whether everyone will actually use it, and whether the coder and front desk can work from the same notes. | Scribeable's Practice plan is $139 a month per clinician with staff seats at no charge. Try it free for 30 days first. |
| Hospital-employed | Whether your organization already licenses one. | Start with what your hospital provides. A personal account that moves with you is a separate decision. |
| Resident or fellow | Your program's policy. | Ask your program director first. Scribeable Pro is free for a year for verified trainees, at 150 notes a month. |
Our ranking of AI medical scribes · What AI scribes cost · Patient consent · How the note is checked
Common questions
Are AI medical scribes accurate?
They make mistakes, and so do people: one peer-reviewed audit found hallucinated content in 31% of AI-generated notes and 20% of physician-written notes. Scribeable does not publish an accuracy number. It links each sentence drawn from the visit or the chart to where it came from, so you can check the line before you sign.
Do AI scribes make up information in notes?
They can, so you read every note before you sign it. Scribeable asks a short question when the plan needs a detail nobody said, such as a dose. Exam systems not discussed in the visit are filled from your own normal exam, anything said in the visit replaces its line, and you review it before you sign.
How much time do AI scribes save doctors?
Published results vary: 0.9 fewer hours of after-hours documentation in one study, and 1.21 fewer minutes of note-active time per encounter with no change in after-hours note closure in another. Scribeable does not publish a time-saved number. Measure your own before and after.
Is an AI scribe worth it for a solo practice?
It is worth testing, because the test is cheap. Scribeable starts with a 14-day Pro trial, no credit card required, and then a Free plan with no monthly note cap once you add your NPI. Pro is $79 a month if you want codes, letters and calculators.
Is an AI scribe worth it for a 5-provider practice?
It is if your clinicians will use it and your coder can work from the notes. Scribeable's Practice plan is $139 a month per clinician, so five clinicians pay $695 a month with staff seats at no charge. You can try the whole plan free for 30 days first.
What are the downsides of AI medical scribes?
You still review every note, a wrong line can get through if you skim, notes can grow longer, and studies disagree on how much time comes back. Scribeable's answer is to make the review faster by linking each sentence to its source, but you still read and sign.
What are the risks of rolling out an AI scribe across a practice?
The usual ones are a missing BAA, unclear patient consent, clinicians who stop using it, and notes the coder cannot rely on. Scribeable includes a signed BAA on every plan and gives each clinician their own templates, and its Practice plan pilot lets you try it for 30 days before you sign the agreement.
How do doctors keep up with charting without staying late?
Close notes between patients, say the plan out loud in the room, use a template per visit type, and move refills and callbacks off the clinician. A scribe such as Scribeable handles the writing; it does not touch your inbox.