Compliance

The AI is advisory — every role

The short, all-staff version: what the software drafts, what only a person can do, and the one sentence to give a patient who asks.

6 min · All staff · lesson v1

Everyone in the practice gets asked about this — usually by a patient, sometimes by a family member, occasionally by a surveyor. This is the short version for all staff. Clinicians have a longer lesson of their own.

The rule, in one line

The software drafts. A person decides.

Nothing clinical happens because the AI concluded something.

What that looks like on screen

You will see the same idea stated in several places, and these are not marketing lines — they are what the product actually enforces:

WhereWhat it says
The differentialFor clinician review only — not a diagnosis. Ranked most → least likely.
OrdersAll orders are DRAFTS… clinician must Confirm
CodingSuggested only — preview. Attest at sign-off; nothing is billed automatically.
The noteYour note — unsigned, and nothing writes into it on its own.
EvidenceNo image or signal is interpreted.
The assistantAI can make mistakes — verify anything clinical.

What a clinician signs

At the end of a visit the clinician ticks five confirmations before anything is final:

  1. I reviewed and edited the draft note.
  2. I reviewed the differential (clinician-review-only).
  3. I reviewed every draft order; none are auto-transmitted.
  4. I reviewed the suggested coding and attest to the codes I accept.
  5. I am the licensed clinician of record for this encounter.

That is a person taking responsibility, in writing, for each thing the software helped with.

The sentence for a patient

"No — the system listens and drafts, and your clinician reviews and decides everything. It doesn't read scans and it doesn't make the call."

And on recording, which is the other question you'll get:

"We're not recording you. It writes down what we say as we say it, and the audio itself is never stored."

Both are accurate. Neither is a softened version.

The record

Every AI action is audited like a human's, with the actor marked as an agent, under the same tamper-evident chain. There is no path by which something happens to a chart and nobody can tell what caused it.

What usually goes wrong

Saying "the computer decided". It didn't, and saying so undersells your colleagues and misleads the patient.

Softening the recording answer. "It's just for notes" invites a follow-up you can't answer. Say plainly that the audio is never stored.

Treating the disclaimers as boilerplate. They describe what the software will and won't do.

Guessing when a patient presses. "I'll get you a proper answer" is always available, and better than an improvised one.

Check yourself

No score, no account — just make sure you can answer these before you move on.

  1. A patient asks "is a computer deciding my treatment?" What do you say?

    Show the answer

    No. The software listens, drafts and suggests; a clinician reviews everything and decides. Orders stay drafts until a clinician confirms them, the differential is labelled not a diagnosis, and the clinician signs five separate confirmations at the end of the visit taking responsibility for the note, the differential, the orders and the codes.

  2. Does the AI read X-rays?

    Show the answer

    No. The product says so in its own evidence panel — No image or signal is interpreted. Interpreting a medical image is a regulated category of software this product deliberately stays out of. It shows images; a clinician reads them.

  3. An AI suggestion turns out to have been wrong. Is there a record of it?

    Show the answer

    Yes. Every AI draft is stored as a draft-only, audited agent action — This AI draft is persisted as a draft-only, audited agent action — with its own tamper-evident audit trail. What was suggested, and what the human did with it, is recoverable.