A practical method for practice administrators to compare virtual human medical scribe services, ambient AI tools, and the possibility of using both together.
When assessing the cost of documentation support, most practices begin by looking at two figures: the monthly fee for an AI subscription and the hourly rate for a scribe. These costs are easy to determine and obviously play a role in the decision. It is more difficult to see how much time each method actually gives back to the doctor who has to check the note before it is signed, and what takes place on the day that the tool or the person is not available. Ambient AI has indeed made real progress and is used well by many clinics. For the majority of practices, the issue is not whether AI should be present in the examination room, but rather whether AI by itself is sufficient to get the note to where it needs to be.
Key Takeaways
- Measure the minutes, not the promise. The largest controlled study to date found ambient AI scribes saved about 16 minutes of documentation time per eight hours of patient care, with bigger gains for frequent users.
- Ask who reviews the note before the physician signs it. AI drafts still need a careful read, and that review time belongs in any cost comparison.
- Check usage by physician. A tool used in a third of visits delivers roughly a third of the benefit.
- Plan for the day the tool or the person is out. Coverage matters as much as capability.
- Count the full cost per physician, including setup, review time, and unused licenses.
- Consider pairing a trained human scribe with the AI tool your physicians already use, so they receive a reviewed, completed note instead of a draft to fix.
What should a clinic ask about any documentation option?
Run these with your physicians and with every option on the table: an in-house scribe, an AI tool built into the EHR, a third-party AI scribe, an outside human scribe service, or a combination. Same questions, every option.
- How many minutes each day does each physician actually receive when the time is measured rather than estimated?
- Who reviews the note before the physician signs it, and how long does that review take?
- How many of our physicians use it in most of their visits?
- What happens on a day it is not available?
- How well does it handle our specialty and our most complex visits?
- How much does it cost per physician when you take into account the time it takes to set it up and review it?
It is useful to know this at the beginning if you are unable to answer some of the questions; typically, a short pilot involving one or two doctors will do so within a few weeks.
Ask your present configuration and every new option the same six questions, and the comparison becomes fair.
How much actual time do ambient AI scribes save?
The time savings are genuine and deserve to be taken seriously; in a randomized trial carried out at UW Health, ambient AI was associated with a substantial reduction in burnout scores and about 30 fewer minutes of documentation each day per provider.
Other larger studies indicate smaller average figures. A study involving multiple sites, which was published in JAMA in April 2026, compared more than 1,800 clinicians who used AI scribes with 6,770 clinicians who did not. The use of AI scribes was linked to 13 fewer minutes of total EHR time and 16 fewer minutes of documentation time over an eight-hour period of scheduled care, representing reductions of 3% and 10% respectively. There was no significant difference in after-hours EHR time between the two groups. A randomized trial carried out at UCLA involving two AI scribes and 238 physicians found that one of the products reduced the time spent in the notes by 9.5%, while the other showed no significant change, even though both slightly improved the burnout measures.
Scribes themselves have a record to show in this regard. In a crossover study carried out by Kaiser Permanente and published in JAMA Internal Medicine, general practice doctors found that they carried out considerably less after-hours documentation when a scribe was with them and spent a greater portion of each visit on the patient rather than on the computer.
A useful test: for two weeks, have each physician note when the last chart of the day is closed. If charts close before the physician leaves the building, the current workflow is doing its job. If they close at 9 p.m., that is the number worth improving, and any option should show how it moves it.
A few minutes per visit adds up. Still, the physician finishing charts at home is the problem most practices are trying to solve, so after-hours time is the measure to watch.
Who checks the note before the physician signs it?
Every option ends the same way: the physician reviews and signs the note. That does not change with a human scribe or an AI scribe. What changes is how much the physician has to fix.
A pilot study carried out at UC Davis and published in JMIR Medical Informatics had physicians assess 356 AI-generated clinical notes. In 18% of the cases there were accidental omissions, in 11.5% hallucinations occurred, and in 9.3% accidental inclusions were found. Most of the errors were minor and approximately 95% of the notes had no significant errors, although 5.3% of them contained errors which, if not corrected, would pose a serious risk. The amount of editing done by physicians varied greatly, ranging from about 2% to 69% of the AI-generated words. An evaluation by the Veterans Health Administration, which appeared in the Annals of Internal Medicine in April 2026, showed that notes written by clinicians scored better than those produced by 11 AI scribe tools in all quality areas, the authors suggesting that AI output should be treated as a draft requiring review.
The researchers also tested the human scribe model in the same way. In the first randomized trial of medical scribes, which was published in the Annals of Family Medicine, the scribes prepared the documentation and then the doctors reviewed it before signing their names. The doctors gave higher ratings regarding the quality and accuracy of the charts, and a greater proportion of the charts were finished within 48 hours.
Here’s something to consider: ask every physician who uses an AI tool to time themselves as they review five recent notes, then record the changes they make. In this situation, if the review takes one minute and the edits are small, then the tool is serving that physician well. But if the review takes five minutes and the physician catches missing findings, then those five minutes count as part of the actual cost.
Review time is the hidden line item in every AI scribe budget.
Are your doctors really employing the tool?
Many physicians are willing to try AI documentation, and many like it. Usage determines the results.
In the JAMA multisite study, clinicians who used AI scribes in more than half of their visits saw about a 2x reduction in EHR time and a 3x reduction in documentation time. Only 32% of users reached that level.
A useful test: pull usage by physician for the last 30 days, which most AI tools can report. If most physicians use it in most visits, the subscription is earning its keep. If use is uneven, it is worth asking why: specialty fit, note format, editing time, or patients who prefer not to be recorded.
It is not a matter of whether doctors are open to using technology; different doctors have different ways of documenting, and a tool that works well for one may not work well for another. Instead of asking each doctor to adjust to the tool, a human scribe adjusts themselves to each doctor.
What happens when the scribe or software isn’t there?
An in-house scribe who knows a physician’s style is a real advantage. So is software that never calls in sick. Each still has gaps worth planning for.
The scribes who work in-house do go on vacation, fall ill, and then leave the job, and the position is usually a step that people take as they go on to attend school or pursue another career. Software can also experience outages, struggle in noisy environments, and relies on patients agreeing to have themselves recorded. In such cases, if either the scribes or the software is not available, the doctor completes the charting.
A good way to check this is to count the number of clinic days in the most recent quarter on which a doctor recorded notes without the usual support. If the answer is zero, then the problem isn’t coverage; but if the figure is ten or more, the practice is already covering for the lack of physician time.
It is worth asking about coverage when considering each option before you actually need it.
Does this align with your area of specialty and your most difficult visits?
With clear-cut visits, a great many AI programs generate clean drafts; it is in the area of complexity that the range of options begins to differ.
In the JAMA multisite study, primary care clinicians achieved the greatest time savings, whereas the differences for the surgical specialties were not statistically significant. The VA evaluation deliberately incorporated visits which involved accented speech, background noise, and masked speakers, since these are typical conditions in real clinics and more difficult for software.
A useful test: pick the five visit types that take your physicians the longest to document, such as a new spine patient or a multi-problem follow-up, and ask each option to show how it handles those, not a demonstration visit. Specialty-trained human scribes learn the terminology, templates, and preferences of a specific physician, which matters most in the visits AI finds hardest. If you are in orthopedics, this review of orthopedic medical transcription might be interesting.
How much does each option actually cost per physician?
AI subscriptions often appear cheaper on paper, and in some cases they are actually cheaper. Prices vary a great deal, ranging from low-cost self-serve tools to enterprise products which are priced by quote and include implementation fees. Human scribe services are usually charged on an hourly basis for the hours that the office schedules.
The subscription involves just a single line; when making a fair comparison, it is necessary to take into account the time spent by the physician looking over and correcting the drafts and the licenses that are not used. Regarding revenue, the JAMA multisite study found that the additional visits associated with the use of AI scribes were statistically significant but only amounted to about $167 per month per adopting clinician.
A good way to do this is to prepare one-page comparisons for each physician, including four lines that list the monthly cost, the number of minutes saved each day (based on actual measurements), the number of minutes spent on reviews each day, and the proportion of visits that are covered. Just because the lowest subscription price is offered doesn’t mean it has the lowest cost per completed note.
Is it possible for a human scribe to work with the AI tool that you currently have?
In the case of many practices, the decision isn’t between human and AI; rather, it’s a question of whether or not to have a trained person work alongside the tools that physicians currently use.
In the case of a paired model, the AI records the encounter and prepares the first draft of the note. A trained scribe then looks at that draft, fills in any omissions, corrects any errors, applies the physician’s template and specialty guidelines, and inputs the note into the EHR. The physician then reviews the note and signs it rather than editing the original draft.
What AI Alone Leaves to the Physician
The Paired Model: AI and a Trained Scribe
→
→
→
Sources: JAMA multisite study, April 2026 (16 minutes, 32%); UC Davis pilot, JMIR Medical Informatics, 2026 (18%).
It is here that the return is the strongest. The practice continues to keep its investment in AI. Physicians allocate less of their own time to the review process. There is a human check placed between the draft and the signature, that is where the omissions and errors mentioned above are detected. Those doctors who have stopped using the AI tool since the cleaning took too long now have a reason to use it again.
Introducing a scribe won’t make an ill-suited AI tool suitable for a particular specialty, and relying solely on AI will not eliminate the review stage; it just places each element in the role where it can perform best.
The physician always maintains his clinical authority. The scribe records everything that the physician says and does, while the physician himself makes all the clinical decisions and signs the note.
Useful to ask your physicians
- Which visits does it take the longest to document today and why?
- How long does it take you to look over a note when you’re using an AI tool, and what kind of changes do you typically make?
- Between reviewing a draft and reviewing a completed note, which would you prefer?
- What would you prefer a scribe to do: stay with you during the visit or take your dictation afterwards?
Curious how a trained scribe would work alongside your physicians’ AI tool? Let’s start with one physician.
Where does this leave your practice?
Ambient AI is a useful tool, and it keeps improving. Human medical scribe services for clinics remain a strong option because they cover what AI still finds hard: careful review, complete notes, specialty nuance, and consistent use from one physician to the next. For many practices, the best result comes from using both.
Running the six questions with one or two physicians over a few weeks gives the practice its own numbers, not vendor estimates. That is a productive place to land, whichever option comes out ahead.
DataMatrix Medical has supported medical practices for more than 25 years, with a US-based team serving 300+ practices across 35+ specialties. Its specialty-trained virtual scribes work live or offline, enter notes directly into any EHR, and include a backup scribe; no long-term contracts; 30-day cancellation notice.
References
- Mishuris RG, Rotenstein LS, et al. “Changes in Clinician Time Expenditure and Visit Quantity With Adoption of Artificial Intelligence-Powered Scribes: A Multisite Study.” JAMA, April 2026. DOI 10.1001/jama.2026.2253. Summary via Mass General Brigham: massgeneralbrigham.org
- Lukac PJ, Mafi JN, et al. Randomized trial of two ambient AI scribes (UCLA). NEJM AI, November 2025. Summary via HealthDay: healthday.com
- Afshar M, et al. “A Pragmatic Randomized Controlled Trial of Ambient Artificial Intelligence to Improve Health Practitioner Well-Being.” NEJM AI, 2025;2(12). UW School of Medicine and Public Health summary: medicine.wisc.edu
- Taylor SL, et al. “Quality of Clinical Notes Created by Ambient Listening Generative AI: Pragmatic Prospective Pilot Study.” JMIR Medical Informatics, 2026. DOI 10.2196/86474. escholarship.org
- Veterans Health Administration evaluation of 11 ambient AI scribes vs. clinician-written notes. Annals of Internal Medicine, April 2026. Summary via Becker’s Hospital Review: beckershospitalreview.com
- Gidwani R, et al. “Impact of Scribes on Physician Satisfaction, Patient Satisfaction, and Charting Efficiency: A Randomized Controlled Trial.” Annals of Family Medicine, September 2017. DOI 10.1370/afm.2122. annfammed.org
- Mishra P, Kiang JC, Grant RW. “Association of Medical Scribes in Primary Care With Physician Workflow and Patient Experience.” JAMA Internal Medicine, 2018. pmc.ncbi.nlm.nih.gov
Nathaniel Smathers is the VP of Client Education and Marketing. He is also a long time contributor of the DataMatrix Medical blog and has a background in healthcare content creation for over a decade. Nathaniel is passionate about exploring the intersections of healthcare, data analysis, and digital innovation.
