Artificial intelligence
Set the models and prompts behind the AI SOAP draft and the Ask AI chart assistant your providers use.
Artificial intelligence is the settings tab where a Provider Network Admin picks the models and writes the prompts behind two things providers use: the AI-drafted SOAP note and the Ask AI chart assistant.
Questions people ask about this page
Does the AI write into a patient chart on its own?
No. Nothing here is automatic. A provider has to click Generate Notes in the Evaluation step to produce a SOAP draft, and the draft lands in four editable boxes — Subjective, Objective, Assessment, Plan — that the provider edits before moving on. Before recording a decision the provider signs an attestation that reads I reviewed all AI-assisted output, verified its accuracy, and exercised independent clinical judgment as the provider of record.
What can Ask AI actually see?
Only that one patient's chart — intake answers, submitted forms, order history, prior visit documentation, sync-visit history, and labs. It is not a web search and it cannot see another patient. Providers open it from the Ask AI button in the charting footer; the panel is headed Ask AI with the subtitle Patient chart assistant.
A prompt field is blank. Is AI switched off for that section?
No. Blank means the curated default is in use — the badge reads Default active for the two system prompts and Default for the rest. There is no off switch on this page. To stop providers seeing an AI-drafted evaluation at all, turn off Evaluation (SOAP) in Settings → Operations.
Why can't I choose my own model?
Only curated Gemini and OpenAI models are offered. The dialog lists seven; there is no free-text model field.
When does a change take effect?
Immediately, for the next generation. The prompt dialog says so: Save applies immediately to new SOAP generations. Notes already generated are untouched. Announce material prompt changes before a clinical shift.
Does Test send anything to a real patient?
No. Test runs a generation against synthetic sample patient data using this network's current configuration and shows the result in a dialog. Nothing is saved to a chart.
Two cards say "model". Which one do providers notice?
SOAP model drafts the four SOAP sections. Visit assistant answers questions in the Ask AI panel. They are set separately and can be different models.
Where to find it
Left menu → Settings → the Artificial intelligence card. Direct link: /settings/ai. Visible to: Provider Network Admin only.
The card on the Settings index reads Artificial intelligence — Adjust the AI note-generation prompts and configuration used by the network. The card inside is titled AI Configuration with the subtitle Model + prompt controls for SOAP generation and Ask AI.
What is on this page
| What you see | What it does |
|---|---|
| Test (card header) | Opens Test SOAP Prompt and generates a sample note from synthetic data. |
| Help icon, About AI prompt controls | Opens a popover explaining how each prompt layer is applied and that only curated models are offered. |
| SOAP model | Opens the SOAP model dialog. Shows Model used for all SOAP sections. Current: {model}. |
| Visit assistant | Opens the Visit assistant model dialog. Shows Model used by Ask AI during visit charting. Current: {model}. |
| SOAP system prompt | Global system-level SOAP instructions, applied across all four sections. |
| Visit assistant prompt | System prompt used by Ask AI inside visit charting. |
| SOAP user prompt | Global user-level SOAP instructions, applied across all four sections. |
| Subjective section prompt | Applies only to the Subjective section. |
| Objective section prompt | Applies only to the Objective section. |
| Assessment section prompt | Applies only to the Assessment section. |
| Plan section prompt | Applies only to the Plan section. |
How to change a model
Open the card
Click SOAP model (or Visit assistant). A dialog opens with the description Choose the model used for SOAP generation across all sections. or Choose the model used by Ask AI in visit charting.
Read what is set now
Under Choose a model the line reads Current selection: {summary}. An unset network reads Default · Gemini 3.5 Flash.
Pick one
Click a model tile. Each shows the provider, the name, and a one-line description.
Save
Click Save model. Closing the dialog with Cancel discards the choice. A card showing the badge Unsaved has a pending change.
| Provider | Model | Description shown |
|---|---|---|
| GEMINI | Gemini 3.5 Flash | Latest Flash model for balanced speed, cost, and quality. |
| GEMINI | Gemini 3.1 Pro Preview | Higher-reasoning Gemini model for complex clinical context. |
| GEMINI | Gemini 3.1 Flash Lite | Low-latency Gemini option for high-volume routine work. |
| OPENAI | GPT-5.5 | Latest standard OpenAI model; excludes Pro-tier pricing. |
| OPENAI | GPT-5.4 | Balanced OpenAI model for quality-sensitive work. |
| OPENAI | GPT-5.4 Mini | Lower-cost OpenAI option for routine generation. |
| OPENAI | GPT-5.4 Nano | Lowest-cost OpenAI option for simple tasks. |
How to edit a prompt
Open the prompt card
Click any of the seven prompt cards. The dialog is titled with the prompt name and described Edit prompt text for this control. Save applies immediately to new SOAP generations.
Read the active default first
For SOAP system prompt and Visit assistant prompt, a blank draft shows the default in a read-only box under No custom system prompt is set. Default system prompt from System/AI_PROMPTS is active: or No custom assistant prompt is set. Built-in visit assistant prompt is active:. Read it before you replace it.
Write the override
Type into Custom Prompt. The placeholder is Leave blank to use default behavior. Keep section prompts specific and do not repeat the global prompt.
Save or revert
Click Save Prompt. Use Default clears the draft so the curated default takes over again. Cancel closes without saving.
Prompts stack. SOAP system prompt and SOAP user prompt apply to every section; the four section prompts apply only to their own letter. Put the house style in the global prompts and keep section prompts to what is genuinely section-specific.
How to test before you roll out
Run it
Click Test in the card header. The dialog Test SOAP Prompt opens: Generates SOAP output using synthetic sample patient data and this network's current prompt configuration.
Wait
It shows Generating test SOAP output..., then a header reading Model: {label} and Generated using sample patient data.
Read all four sections
Check Subjective, Objective, Assessment and Plan for clinical restraint, tone, structure, and length. An empty section reads No output returned. Read any Warnings list.
Iterate
Change one prompt, then click Run Again. Close with Close.
Tell the clinical team
Saved changes apply to the next generation. Announce material changes before a shift starts.
What a provider actually sees
| Where | What the AI does | Who reviews it |
|---|---|---|
| Evaluation step → Generate Notes | Drafts all four SOAP sections in one pass using the SOAP model and your SOAP prompts. | The provider edits the four rich-text boxes directly. Nothing is submitted by the AI. |
| Each SOAP section → Regenerate section | Redrafts that one section using its section prompt. | Same — it replaces text in an editable box. |
| Ask AI panel in the charting footer | Answers questions about this patient's chart using the Visit assistant model and the Visit assistant prompt. Placeholder: Ask about this patient's intake, history, orders, or transcript... | Nothing it says enters the chart. The provider decides what to write. |
| Attestation / decision step | Nothing. The AI has no role in the decision. | The provider signs I reviewed all AI-assisted output, verified its accuracy, and exercised independent clinical judgment as the provider of record. |
Every setting on this page
| Setting | What it controls | Default |
|---|---|---|
| SOAP model | The model that drafts all four SOAP sections. | Default · Gemini 3.5 Flash |
| Visit assistant | The model behind the Ask AI panel in visit charting. | Default · Gemini 3.5 Flash |
| SOAP system prompt | System-layer instructions applied to every SOAP section. | Blank — curated default active |
| Visit assistant prompt | System prompt for the Ask AI conversation. | Blank — built-in default active |
| SOAP user prompt | User-layer instructions applied to every SOAP section. | Blank |
| Subjective section prompt | Subjective only. | Blank |
| Objective section prompt | Objective only. | Blank |
| Assessment section prompt | Assessment only. | Blank |
| Plan section prompt | Plan only. | Blank |
Badges you will see here
| Badge | What it means | What to do |
|---|---|---|
| Custom | This prompt has been overridden by your network. | Open it and read the text before you assume the platform default is running. |
| Default active | A system prompt (SOAP system prompt or Visit assistant prompt) is blank, so the curated default is in use. | Nothing. Blank does not mean off. |
| Default | A user or section prompt is blank. | Nothing. |
| Unsaved | A model card has a pending change that has not been saved. | Reopen the card and click Save model, or reload to discard. |
| Default · Gemini 3.5 Flash | No model has been chosen; the platform default is running. | Nothing. |
| Custom model configured | A stored model value the screen does not recognise. | Open the card and pick one of the seven curated models. |
What can go wrong
| What you see | Why | Fix |
|---|---|---|
| Unable to run prompt test | The test generation failed. | Click Run Again. If it keeps failing, revert your last prompt with Use Default and retest. |
| Provider network context is required to test prompts | Your account is not resolving a provider network. | Sign out and back in; if it persists, raise a ticket from Support. |
| Unable to save AI configuration. Please try again. | The write failed. | Reopen the dialog and save again. The card keeps the Unsaved badge until it lands. |
| Providers say the notes changed tone overnight. | A prompt was saved and applies immediately to new generations. | Open the card, read the custom text, and use Use Default to revert. |
| A section comes back empty in Test. | That section's prompt contradicts the global prompt, or the model returned nothing. | Clear the section prompt with Use Default and run again to isolate which layer is at fault. |
| You want AI off for this network. | There is no off switch on this page. | Turn off Evaluation (SOAP) in Settings → Operations. Providers keep the Ask AI panel; it is not gated here. |
Next
Was this helpful?