How DAX Interprets Instructions
The rules behind building an effective DAX prompt.
DAX Copilot doesn't guess your style. It follows instructions literally. Understanding how it reads those instructions is the difference between generic output and documentation that sounds like you.
What DAX Copilot actually does
It listens. Then it writes.
That's it.
What it does:
- ✓Ambient listening during patient encounters
- ✓Converts conversation into structured clinical notes
- ✓Generates HPI, Physical Exam, A&P, Results
- ✓Accepts custom instruction prompts that control output style
What it doesn't do (yet):
- ✗Cannot pull labs, imaging, or prior visit data
- ✗Cannot access the chart — only ambient listening
- ✗Does not infer clinical reasoning you didn't say aloud
DAX follows instructions literally.
Vague instructions → generic output.
Precise instructions → aligned output.
DAX doesn't know your preferred note structure. It doesn't know you like bullet points over paragraphs, or that you want a ranked differential, or that you always include return precautions at the end.
It only knows what you tell it in the instruction prompt.
Small formatting changes in the instruction layer produce large differences in output. This is why the rules matter.
The Formatting Rules
DAX prompts use four types of formatting. Each one tells DAX something different about what to do with that text.
Plain text → Section headers & labels
DAX prints these exactly as written.
Assessment: ED Interventions: Discharge Plan: Disposition:
"Double quotes" → Exact canned text
DAX prints this word-for-word. Use for standard phrases you always want included.
"Risks and benefits have been discussed with the patient." "Return to the ED for worsening symptoms or new concerns."
[Square brackets] → AI-filled content
DAX fills these in from the conversation. The text inside guides what it listens for.
[1–2 sentences summarizing key findings that support the working diagnosis.] [Most likely diagnosis + 1 line rationale.] [Medication name + brief rationale]
{Curly brackets} → Logic & rules
DAX reads these as instructions but does NOT print them. Use for deletion rules, loops, and constraints.
{Delete if none.}
{Delete any unused lines.}
{Repeat for each problem.}
{Do not include treatments in this section.}
{Insert a blank line before each header.}Quick Reference
| Format | What DAX Does | Example |
|---|---|---|
| Plain text | Prints exactly as written | Assessment: |
| "Quotes" | Prints word-for-word | "Risks discussed." |
| [Brackets] | AI fills from conversation | [Key findings] |
| {Curly} | Rules — never printed | {Delete if none.} |
Principles for strong prompts
These patterns consistently produce better output.
Be explicit about structure
Don't assume DAX knows you want bullets or numbered lists. Tell it.
Separate sections with blank lines
DAX uses blank lines to distinguish between sections. Without them, content bleeds together.
Always include deletion rules
"{Delete if none.}" and "{Delete unused lines.}" prevent template clutter in the final note.
Control sentence length
"1–2 sentences" or "3–6 sentences" gives DAX a concrete constraint. Without it, output tends toward verbose.
Say what NOT to include
"{Do not include treatments in this section.}" prevents content from appearing in the wrong place.
Use loops for repeating content
"{Repeat for each problem.}" tells DAX to expand the template for multiple diagnoses.
What this means for you
You don't need to be technical. You need to be specific.
The formatting rules above are the entire language DAX understands. Four symbols. That's it.
When you use them consistently, DAX produces notes that reflect your reasoning, your structure, your voice.
The styles and building blocks on this site already follow these rules. You can use them as-is or customize them knowing how the system works.