Documentation has quietly become one of the heaviest burdens in clinical practice. The often-cited work by Sinsky et al. (Annals of Internal Medicine, 2016) found that physicians spent roughly two hours on EHR and desk work for every hour of direct patient care. It is no surprise, then, that scribes, both human and increasingly AI, have become a serious part of the conversation about reclaiming clinician time.
The question is not which is universally better. It is which fits your volume, your budget, and your workflow. This is an honest look at both, including where each genuinely falls short.
What a human scribe does well
A trained human scribe is, at their best, an extension of your clinical thinking. They sit in on the encounter, follow the reasoning, and learn your documentation preferences over weeks of working alongside you. That contextual understanding is hard to overstate. A good scribe knows when an offhand comment is clinically relevant and when it is small talk, and they can capture the nuance of a complex visit in a way that reflects how you actually practice.
They can also be trained to a specific clinician. Over time, a scribe internalizes your phrasing, your specialty's conventions, and the way you like your notes structured. For high-acuity or highly individual practices, that adaptation is real value.
Where human scribes get hard
The strengths come with practical costs that scale poorly:
- Cost. A human scribe is a salaried or contracted person. For many smaller practices, that is simply out of reach.
- Scheduling and coverage. Vacation, illness, and turnover mean gaps. When your scribe is out, you are back to documenting yourself.
- Training time. New scribes take weeks to months to reach full productivity, and that ramp repeats with every replacement.
- A third person in the room. Some patients are less candid with an extra person present, which can affect rapport and the privacy of sensitive conversations.
What an AI scribe does well, and where it does not
An AI scribe trades human judgment for speed, scale, and consistency. It is available 24/7, produces a draft almost immediately, never calls in sick, and carries a much lower marginal cost per note once it is set up. For a clinician seeing a high volume of fairly routine visits, that consistency and availability can be the deciding factor. It also keeps the encounter to two people in the room.
The honest caveats matter just as much. AI transcription and drafting accuracy varies with audio quality, accents, background noise, and how unusual the case is. An AI scribe does not truly understand the patient the way an experienced human does, which means every note requires clinician review before it enters the record. And because these tools process audio and clinical text, they depend on sound data handling and clear disclosure of which sub-processors touch your data.
Shifaa AI is one example of a mobile AI scribe built with those caveats in mind. Its voice-to-SOAP scribe uses OpenAI Whisper for multilingual transcription and Anthropic's Claude to draft a structured note, and it deliberately fills empty fields only, never overwriting what the clinician has already written. The doctor reviews and signs off; the tool drafts, it does not decide. You can see the capture-to-review sequence on the how it works page.
The honest takeaway
Neither option is strictly better. A human scribe rewards practices that can absorb the cost and value deep, trained-to-you context. An AI scribe rewards practices that need instant, consistent, scalable documentation at low marginal cost and are comfortable owning the review step. Match the tool to your volume, budget, and workflow, not to the hype.
If you do choose an AI scribe, treat the review step as non-negotiable rather than a formality. The draft is a starting point that saves typing, not a finished medical record. Used that way, with you firmly in the loop, an AI scribe can give back a meaningful share of the documentation time that clinical work quietly takes away.