Ask a clinician which AI mistake they fear, and it usually isn’t the obvious one. A garbled sentence is easy to catch. The dangerous error is the confident one: a clean, plausible line that reads like every other line, sits in the right section, and gives you no reason to look twice. That is the error you sign without noticing.
Akoua is designed around that single problem. Instead of smoothing every rough edge into fluent prose, it shows its work.
Fluent is not the same as correct
An AI note that sounds sure of itself is easy to build. A model can always produce a confident sentence, whether or not the words were actually said. The hard part, and the part that matters in a medical record, is telling you where that confidence is earned and where it isn’t. A fluent guess hides that line. Akoua draws it.
Three ways it flags what it doesn’t know
1. “Not sure I heard this”
When Akoua’s confidence in a word or phrase is low, it doesn’t quietly pick the most likely option. It marks the text with a dotted underline. Tap it and you listen back to that exact slice of audio, so the decision is yours and it takes seconds.
2. “Heard it, couldn’t place it”
Sometimes the words are clear but the meaning won’t sit cleanly in a section of the note. Rather than forcing it somewhere and hoping, Akoua keeps it as a highlighted block with a review tag. Nothing is guessed into a structure it doesn’t belong in.
3. [not stated]
This is the quiet one that does the most work. Anything that wasn’t said is written as [not stated], never invented to make the note look complete. If the examination wasn’t described, the note says so. An empty field is information; a plausible fabrication is a hazard.
Every line is grounded to the transcript
Beyond the flags, each statement Akoua drafts is grounded back to what was actually said in the consult. The note is not a free-form summary that drifts from the conversation. It is tied to the transcript underneath it, so what you review is what was heard.
You confirm before anything is saved
Nothing Akoua drafts is saved or exported until you confirm it. The draft is a proposal, not a record. You read it, resolve the flags, correct anything that needs it, and only then does it become a note. You clear the flags first, so you always know how much is still waiting on your judgement before you commit.
Why “shows its work” beats a fluent guess
A scribe that hides its uncertainty asks you to trust it completely or not at all. Neither is a good place for a clinician to be. By marking what it isn’t sure of, Akoua turns review from a proofread of confident prose into something faster and safer: a short list of exactly the places that need a human. You are not hunting for the one wrong line in a wall of plausible text. Akoua has already pointed at it.
That is also why we are careful about what Akoua is. It is a documentation aid, not a diagnostic device. It doesn’t reason about the patient; it records the consultation honestly and hands you the decisions. The point was never a note that looks finished. It was a note you can sign, because you can see exactly what stands behind every line.