AI clinical reasoning
An optional AI panel that helps you think through a case — what could explain it, what's missing, and what to consider doing — never a diagnosis, never sent without your say-so.
Once an Owner or Manager turns on AI clinical assistance, an encounter gains two extra sections once it reaches the possible-conditions step: Clinical Reasoning and Management Considerations. Both are generated by an AI provider and both are clearly labelled as suggestions — they sit alongside the rule-based possible-conditions list you already have, never in place of it.
📸 Screenshot needed: the Clinical Reasoning panel open on an encounter, showing both sections.
The Clinical AI Usage add-on
Before any of this can be turned on, your workspace needs an active subscription to the Clinical AI Usage add-on (separate from the Clinical Suite add-on itself, and separate from the general "AI Assistant" add-on elsewhere in the platform). Subscribing grants a daily allowance — one shared pool covering every AI feature described on this page, not five separate budgets to track.
If the add-on isn't subscribed, the AI clinical assistance card explains this and links to Manage add-ons — there is no free baseline usage without it.
Once today's shared allowance is used up, you can buy more tokens. Purchased (or admin-granted) tokens are drawn from only after the day's allowance runs out, roll over from day to day, and expire at the end of the calendar month if unused. Your current token balance and a Buy more tokens button appear on the same card once AI clinical assistance is turned on.
Turning it on
Once the add-on is active, this is still off for every workspace until an Owner or Manager turns it on, from the AI clinical assistance card on the Clinical Encounters page. Turning it on shows exactly what leaves the platform when it's used:
Encounter details — symptoms, vitals, severity, known conditions — are sent to an AI provider to suggest possible conditions and next steps, without the patient's name, MRN, date of birth or contact details. A pharmacist always reviews what will be sent before it goes, and decides what to do with the result.
Anyone can see this card; only an Owner or Manager can change the setting. This is separate from your permission to use Clinical Encounters — it's a workspace-wide decision about whether patient data may go to an external AI provider at all.
What is sent — and what never is
Before every single request, you're shown exactly what will be sent, and asked to confirm. It includes: the encounter type, the presenting complaint and symptoms, severity and duration, vitals, the patient's age (not their date of birth), pregnancy status, known conditions, and the triage/possible-conditions results already computed.
It never includes the patient's name, MRN, date of birth, phone number, address or emergency contacts. If you typed something like a phone number into a free-text field (for example, in "relevant history"), it is automatically replaced with [redacted] before anything is sent.
Clinical Reasoning
Select Ask AI to generate this section. It writes out, in plain language:
- Presenting problem — a one- or two-sentence summary of why the patient is here.
- Relevant findings — the facts that matter, pulled together.
- Possible contributing factors — not only differential diagnoses, but things like adherence or lifestyle factors too.
- Important information still required — specific gaps worth asking about or checking.
- Safety considerations — a short, advisory note. This is not a second safety check: the triage result you already saw is the one thing that can ever stop the workflow, and this note can never change it.
Management Considerations
Underneath, Management Considerations lists general next steps to think about — for example "review medication adherence" or "review recent laboratory results." It never names a specific medicine or dose, and it never decides anything for you. You still make the call on the decision step, exactly as before.
What it is not
- Not a diagnosis, and not artificial intelligence pretending to be a doctor. Everything here is phrased as a possibility to consider, never a fact.
- Not the safety check. Triage and mandatory escalation are computed by fixed rules, before any of this runs, and this panel cannot be reached at all while an escalation is open and undocumented — see The triage safety check and Escalation and overrides.
- Not always available. If your workspace hasn't subscribed to Clinical AI Usage, hasn't turned the assistance toggle on, or has used up today's shared allowance and token balance, you won't see the buttons, or you'll see a plain message explaining why.
Improving the SOAP note with AI
On the final sign step, if AI clinical assistance is on, an Improve with AI button appears next to the drafted note. It rewrites the same facts into cleaner clinical prose — it never adds a fact, finding or recommendation that wasn't already in the draft. You can still edit anything afterward, and you still sign it yourself.
History Summary
On the patient's Dashboard tab, a History Summary (AI) card lets you ask AI to write a short orientation paragraph before you start: known conditions, recent diagnoses, and the outcomes of past visits, in plain language. It never suggests a new diagnosis or treatment — it only summarizes what is already on record.
It needs the same AI opt-in as Clinical Reasoning, plus permission to view the patient's Notes and Diagnosis in addition to the patient record itself — the summary is built from all three. Like the rest of this panel, nothing that identifies the patient is sent, and it counts against its own separate daily usage limit, not Clinical Reasoning's.
Two things are always true of this summary, by design:
- It is never saved. Every click regenerates it fresh from the patient's current record — there's no cached version to go stale as new visits and notes are added.
- Confidential notes are never included, even if you personally have permission to view confidential notes elsewhere in the system. Only non-confidential note titles, settled (signed) encounter outcomes, and recorded diagnoses are ever summarized.
Follow-up Progress Summary
When you start a new follow-up encounter, you can link it to the specific prior signed visit it continues — a dropdown at intake lists the patient's settled visits to choose from. Linking is optional, but only a linked follow-up gets a Follow-up Progress (AI) panel once it reaches the possible-conditions step.
Select Ask AI there to compare today's visit against the linked one: a short recap of the prior visit, what seems to have changed, a plain-language progress note (improved / unchanged / worse — never a diagnosis), and general next steps to consider. Same rules as Clinical Reasoning: additive only, never a second safety check, and it stays available (or blocked) exactly where the ordinary panel does.
There's deliberately no fallback to "the patient's most recent visit" if you don't link one — a patient juggling more than one issue could have the wrong two visits compared, so the panel simply doesn't appear until you link one.
Follow-up completion note draft
When you complete a follow-up task (from Appointments & Follow-ups), an Ask AI button next to the note field drafts a short, generic starting note — reviewed and edited by you before you save it. It never invents a specific clinical finding: if the follow-up came from a clinical encounter, it can mention that visit's original complaint and outcome for context, but the actual "what happened" during the follow-up itself is still yours to write.
Counselling Points
Once a decision has been recorded on an encounter, a Counselling Points (AI) card appears on the sign step. Select Ask AI for a short list of spoken points — general self-care guidance for the condition, and clear signs that mean the patient should come back sooner — for you to relay to the patient yourself, in your own words.
This is never text shown to or sent to the patient directly, and it never names a specific medicine, dose or drug interaction — medication-specific counselling needs real structured drug data and is a separate, not-yet-built feature (see the Medication Management and Drug Information Center phases). What you get here adapts to the visit: an acute complaint gets self-care and red-flag guidance, a chronic disease review gets general self-management talking points (diet, monitoring, adherence) grounded in what was actually reviewed.