The short versionIf approving the AI's note is faster than reading it, the review step is decorative. Either make the draft genuinely reviewable, or stop claiming a human approved it.
The four minutes after the patient leaves
The encounter is over, the patient is in the corridor, and the note is owed. This is the gap every clinical documentation product is sold into, and it is a real one: the further a note drifts from the moment of the conversation, the more of the conversation has already gone.
What a physician needs in those four minutes is not a blank field, and not a finished document either. It is a draft close enough to be worth correcting — because the realistic alternative to a good draft is not a better note, it is a shorter one written at the end of the day.
A note nobody edits is not a note a clinician approved. It is one they gave up on.
Where the loop actually sits
“Human in the loop” is a claim about an interface, not about an intention. In Clinical HealthAI the loop is four specific places in the software, and each one is somewhere a clinician can change the outcome rather than merely acknowledge it.
- The generated note arrives as a draft and cannot leave as anything else until a clinician approves it.
- Clinical terminology is normalised for consistency, but the physician's own phrasing is kept alongside it rather than replaced by it.
- Uncertainty is shown where the model has it, instead of being smoothed into confident prose that reads identically whether it is right or wrong.
- Approval is attributable: who reviewed it, what they changed, and when — kept as a record a person can read, not a log line.
What it does not do
It does not decide anything clinical, and it is not more accurate than the clinician reviewing it — it is faster, which is a different claim and the only one we make. Documentation support does not extend to diagnosis or triage, or to anything that would put the software in the position of holding an opinion about a patient.
We also do not publish an accuracy figure for it. The number that would matter is how it behaves on your specialty, your dictation habits and your templates, and we have not measured it on those. A pilot measures that; a brochure cannot.
Written about Clinical HealthAI
