In a virtual practice, ambient AI scribes cost from $39 to $700 for each clinician every month. As reported by a study of more than 1,800 clinicians in five academic medical centers, those tools lower the time spent on documentation by about 16 minutes for every eight hours of care. It is a reduction of 10%, which differs from the 70% decrease that many sellers suggest. By using self-serve pricing, the cost remains lower than the value of the time saved – but when organizations pay enterprise prices, the investment is only efficient if clinicians use the software for most visits, though only 32% of participants did so. For virtual practices, a second issue exists because recording a video session creates requirements for consent and data storage – those duties are part of the telehealth platform rather than the electronic health record.
What do ambient AI scribes actually cost?
To understand the costs of ambient AI scribes, a comparison of pricing models is necessary.
| Option | Cost per provider | Notes |
|---|---|---|
| Human in-person scribe | $32,000 to $42,000 per year | Monthly cost is roughly $2,667 to $3,500 |
| Virtual scribe service | $1,200 to $4,000 per month | Humans still perform the work |
| AI scribe, self-serve | $39 to $120 per month | Freed AI is $39 – Dragon Medical One is about $99 |
| AI scribe, mid-market | $120 to $300 per month | Heidi Health is $110 to $180 |
| AI scribe, EHR-integrated | $400 to $700+ per month | DAX Copilot is about $600 |
There is a difference of approximately 15 times between the lowest and highest AI price tiers – this range determines if the tool is profitable.
How much documentation time do AI scribes actually save?
By observing 1,800 clinicians and 6,770 control subjects from 2023 to 2025, researchers measured actual time savings. Adopters reduced documentation by 16 minutes and spent 13 fewer minutes in the medical record for every eight hours of work. In relative terms, this is a 10% decrease in documentation and a 3% decrease in total time within the EHR.
And three specific details from the research are important.
- Due to the high impact of frequent use, clinicians who used scribes for over half of their appointments saw triple the documentation savings. Only 32% of users reached that threshold.
- But work performed after standard hours did not change for the clinicians using the tools.
- And revenue increased by only $167 per month for each clinician, which is equal to one extra patient every 14 days.
Is the review tax already in these numbers?
If people ask if those figures include the time spent reviewing notes, the answer is yes. Since the data comes from EHR audit logs, the 16-minute result includes the time clinicians spent reading and editing the text. To be clear, the time required for review is the reason the savings are 16 minutes instead of the higher numbers that sellers advertise.
Where does the error rate land clinically, and who is liable?
On the topic of clinical errors, a review in NPJ Digital Medicine indicates that ambient AI scribes have error rates between 1% and 3%. For comparison older dictation systems have error rates between 7% and 11%. When clinicians use the tools, four types of errors occur.
- Fabrication – the AI might describe physical exams that did not happen.
- Omission – it might leave out information that the doctor and patient discussed.
- Misinterpretation – it might misunderstand statements that depend on specific contexts.
- Speaker attribution errors – it might assign words to the wrong person when multiple people are in the room.
With a 2% error rate, a clinician seeing 20 patients daily will find mistakes in multiple notes every week. Evidence also shows that accuracy is lower for Black patients than for White patients. If a practice serves a diverse population, the organization should request accuracy data for specific groups from the vendor.
In the context of vendor contracts, legal liability resides with the individual clinician. It is standard for those agreements to place clinical responsibility on the person who signs the medical note. Due to their classification as administrative tools instead of medical devices, most ambient scribes operate without FDA oversight. For those reasons, you should establish a documentation policy. In this policy you must define which sections require verification line by line before a signature is applied. And you should determine who performs a monthly audit of samples. To manage errors you must define the procedure for when a mistake enters a medical chart.
What changes when you record a video visit instead of an office visit?
When you record a video visit instead of an office visit, the situation changes. It is at this point that ambient capture is no longer a question of the EHR. If an AI scribe must announce a recording, the requirement is a result of state wiretap laws rather than HIPAA. Paubox notes that a recording of a confidential encounter without the consent that a state requires can create liability for wiretapping – this is true regardless of the completeness of HIPAA documentation. In states that require all-party consent, like California, Illinois, Pennsylvania, Florida, Massachusetts, Washington and Maryland, this rule is important.
Virtual care adds a complication – in a video visit, a patient is often in a different state than the provider. As a result the applicable consent rule is subject to change between appointments. There are four questions to answer before you enable ambient capture on video:
- Is consent recorded and marked with a timestamp at the beginning of the encounter?
- Where is the audio file stored, how long is it kept and who is able to retrieve it?
- Does the BAA include the scribe vendor and prohibit the use of recordings to train models?
- Are you able to produce an audit trail for each encounter that shows the method, the location of the patient and the consent?
To answer questions one, two and four, you must look at platform capabilities rather than EHR capabilities. For practices that evaluate white label telehealth platforms with HIPAA compliance, consent capture and the retention of recordings are now primary selection criteria. By using single-tenant architecture, where each client has dedicated infrastructure, you find that questions about retention and audit trails are easier to answer. On the platform from Healee, there is a history of more than 5 million appointments for more than 200 clinics and 1 million patients under HIPAA and GDPR. And the configurable intake workflows that identify the visit method are also able to capture consent for recording.
Break-even for a 5 to 10 provider practice
For a practice with five to ten providers, you can calculate the break-even point. Consider a practice with six providers on a self-serve tier that costs $89 per provider each month or $6,408 each year. At the average of 16 minutes per eight clinical hours across 184 days, each provider gains approximately 49 hours each year. With a cost of $1,068 per provider, the price is approximately $22 for every hour of time that returns to the provider. The measured $167 per month in new revenue is enough to cover the subscription cost.
If you apply the same math to an enterprise tier that costs $600, the cost is $7,200 per provider. Against 49 hours the cost is roughly $147 per hour – this is a high rate that is equal to or greater than the value of an hour of physician time in most small practices. If a user reduces documentation by three times that amount, which is 147 hours, the cost is roughly $49 per hour. At this rate the cost is defensible – but the enterprise tier is a bet that usage is consistent and only one third of clinicians achieved that level.
There is a case for waiting. It is better to wait if you are unable to ensure providers use the tool for more than 50% of visits. If you operate in states with all-party consent rules without a consent workflow at the platform level, you should also wait. And you should wait if you receive a quote for an enterprise tier without data from a pilot of your own providers. To get an accurate number, a 90-day pilot on a self-serve tier with two providers costs a small amount of money.
Frequently asked questions
How much time do ambient AI scribes save per provider?
On average there is a reduction of 16 minutes of documentation time for every eight hours of patient care – this is a 10% reduction based on data from 1,800 clinicians. For clinicians who use a scribe for more than half of their visits, the reduction is three times larger. In some earlier studies, the reduction was as small as 34 seconds per note.
Are AI scribes less expensive than human scribes?
The cost is substantially lower. A human scribe who is present in person costs between $32,000 and $42,000 per provider each year. In contrast self-serve AI tiers are available starting at $39 per provider each month. A more difficult comparison is between different AI tiers. The enterprise options that integrate with the EHR cost between $400 and $700 per month – those require much higher usage to be justifiable compared to self-serve options that cost between $39 and $120.
Is patient consent necessary to use an AI scribe on a telehealth visit?
In all-party consent states, people are required to agree to the recording before the process begins – those states are California, Illinois, Pennsylvania, Florida, Massachusetts, Washington and Maryland. It is important to note that HIPAA compliance is not equivalent to state wiretap law. For practices that operate in multiple states, the policy that offers the most protection is to obtain explicit consent for every encounter.
Who is responsible if an AI scribe writes something false into a chart?
The clinician who signs the note is the person who is responsible. There are no vendors that take clinical liability and most ambient scribes are not medical devices that the government regulates. Practices are encouraged to have a written policy that states which sections of a note require a person to verify the content before they sign.
Which parts of a clinical note are safe to let AI draft?
Narrative history and plan sections are less likely to cause harm because a person can see errors when they read the text again. By contrast physical examination findings, medication lists and details that a future clinician will use to make decisions require a person to review every line – this is necessary because there are instances where those systems create examination findings that did not occur.
Should a five provider practice adopt an AI scribe now?
As the cost for self-serve options is low compared to the time saved, a trial with two providers is a choice that involves little financial danger – but you should wait to make a commitment for an entire enterprise until you possess data from your own use. And you must decide how to obtain consent for video visits before you start using ambient recording.
If you are evaluating how ambient capture, consent and recording retention are managed within your workflows, you can request a demo to see how the parts are integrated on a platform that uses your branding.
Sources:
- AI Scribes Linked to Modest Reductions in Electronic Health Record Use and Clinical Documentation Time – Mass General Brigham, reporting JAMA study, April 2026
- Large AI scribe study finds modest time savings, inconsistent use – STAT News, 2026
- Beyond human ears: navigating the uncharted risks of AI scribes in clinical practice – NPJ Digital Medicine, 2025
- AI Medical Scribe Pricing 2026: What Practices Pay – Commure, 2026
- Does an AI clinical scribe need to announce that it is recording? – Paubox, 2026