Why your scribe's note doesn't sound like you
The AI scribes are good now. That’s worth saying plainly, because a few years ago it wasn’t true. Tools like DAX Copilot, Abridge, and Nabla will sit in the room with you, listen to a messy, nonlinear human conversation, and hand back a coherent clinical note. As a piece of technology, that’s remarkable, and the physicians using these tools aren’t wrong to like them.
But watch what happens next. The note comes back, and the physician starts editing.
Not correcting errors, mostly — editing voice. Rephrasing the assessment because that’s not how they’d say it. Restructuring the plan because they order things differently. Softening a line here, sharpening one there, deleting the boilerplate sentence the scribe loves that they would never write. The note is accurate and complete and reads like it was written by a very competent stranger. Because it was.
So a meaningful slice of the time the scribe saved gets spent making the note sound like it came from you. Every visit, every note, forever.
Style isn’t a settings menu
The industry’s answer to this has been configuration: templates, preference toggles, “custom vocabularies.” I think that answer misunderstands the problem.
A physician’s documentation style isn’t a set of options. It’s a learned representation — something closer to handwriting than to formatting. When I look at what actually makes a note mine, I see at least five layers:
The words themselves. Every physician has a lexical fingerprint — specific phrases, specific constructions, reached for consistently and unconsciously.
Structural rhythm. How sections flow into each other. What gets its own paragraph. Where the note breathes and where it compresses.
Reasoning density. How much diagnostic thinking gets externalized on the page versus implied. Some physicians write their differential out loud; others document the conclusion and carry the reasoning in their heads. Neither is wrong. They are different notes.
Hedging language. How uncertainty gets documented — “likely,” “cannot exclude,” “will reassess” — is not decoration. It’s clinical judgment made visible, and it carries real medicolegal weight. A tool that flattens your hedging is editing your practice, not just your prose.
Audience calibration. The same physician writes a progress note, a discharge summary, and patient-facing instructions in three deliberately different registers. Style that doesn’t shift across note types isn’t style; it’s a template wearing a costume.
No dropdown menu captures that. Which is why the dropdown menus haven’t.
What I’m building
NoteWright sits downstream of whatever scribe you already use — though strictly speaking, it doesn’t care where the text comes from. Scribe output, dictation, a rough draft you typed between patients: you paste the input in, and it comes back rewritten as your note — your phrasing, your structure, your rhythm, your hedging, calibrated to the note type in front of you.
One thing the input never needs, by design: patient identifiers. Names, dates, record numbers — none of it has style, so none of it contributes anything to the rewrite. Remove identifiers before you paste and the output is identical. A tool that only does the formatting work has no business holding the information that makes a note a patient’s note, and this one is built accordingly.
It is deliberately not a scribe. It doesn’t listen to your encounters, and it isn’t trying to replace the transcription layer — that problem is being solved well by companies with far more microphones than me. It’s also not an EHR integration. It does one thing: it takes a note written by a competent stranger and makes it sound like it was written by you.
Built for one physician first
I should be honest about who this is for right now: me.
I’m a family medicine resident, which means I live inside the documentation problem I just described. I’m also the first user, the test case, and the quality bar. The rule I’ve set for this project is simple — NoteWright has to earn a place in my own workflow before it’s allowed to ask for a place in anyone else’s. If it can’t consistently produce a note I’d sign without wincing, nothing else about it matters.
This journal is where I’ll document the build — the design decisions, the principles behind them, and the places where reality pushes back. (A note on dates: the earliest entries, this one included, were written retrospectively from the project’s own records — specs, commits, decision logs — and are dated to the work they describe. Anything written in the moment says so.) If you’re a physician who’s ever sighed at a scribe note before rewriting half of it, you’re who I’m writing for.
More soon.