Every doctor documents differently. Some want drusen graded on a three-point scale; others want a narrative description. Some insist on specific follow-up language in the plan; others prefer shorthand. Custom instructions let you tell Doctora's AI exactly how to write each section of the clinical note so the output matches your style from the start--not after editing.
What custom instructions do
When Doctora processes a recording, the AI follows a clinical schema that defines every section of the note: chief complaint, anterior segment, posterior segment, assessment, and so on. Custom instructions are doctor-written directives that get injected directly into that schema before the AI runs. They are not suggestions buried in a prompt--they are attached to the specific field they apply to, so the AI sees your instruction at the exact moment it generates that section.
There are roughly 40 configurable fields organized across six categories: Background, Examination, Refraction, Special Testing, Contact Lens, and Assessment. You can instruct any combination of them, or none at all.
Two ways to add instructions
Method 1: The pencil icon (inline)
After Doctora generates a note, every section heading in the editor has a small speech-bubble icon next to it. Click it to open a popover where you can type your instruction, choose a priority level, and save. The instruction takes effect on your next encounter--no need to navigate away.
This is the fastest way to add an instruction the moment you notice the AI doing something you would correct. See a pterygium described as "growth on the conjunctiva" when you want "nasal pterygium, grade II"? Click the icon next to Anterior Segment, type your preference, and it sticks.
When a field has an active instruction, the icon changes to a colored badge showing a truncated preview of your instruction text (blue for tips, amber for important, red for required).
Method 2: Settings bulk editor
Go to Settings > Custom Section Instructions to see all your instructions in one place. Here you can:
- Add new instructions by typing a section name (autocomplete helps you find the right field)
- Edit or delete existing instructions
- Set the global instruction (described below)
- See which fields already have instructions--they appear highlighted in the autocomplete dropdown
The bulk editor is the better choice when you are setting up Doctora for the first time or want to review everything at once.
The three priority levels
Each instruction has a priority that controls how strongly the AI follows it.
Tip (blue) -- A helpful suggestion. Use this for stylistic preferences where occasional deviation is acceptable. The AI treats this as a "NOTE" annotation on the field.
Example: "Prefer bullet points over paragraph format for this section."
Important (amber) -- The AI will prioritize this. Use it for preferences that matter to your clinical workflow but where rare exceptions are understandable. The AI treats this as an "IMPORTANT" annotation.
Example: "Always include lens status (clear, NS 1+, PCIOL, etc.) even when the patient does not mention it."
Required (red) -- Must be included in the output. Use this for non-negotiable formatting or content rules. The AI treats this as a "REQUIRED" annotation and will follow it unless the transcript contains no relevant information at all.
Example: "Never abbreviate medication names. Always use the full generic name."
A good rule of thumb: start with Tip. Promote to Important if you find yourself consistently correcting the output. Reserve Required for documentation standards you truly cannot compromise on.
The global instruction
Above the field-specific instructions in Settings, there is a General AI Preferences text box. Anything you write here applies to every section of every note. Field-specific instructions take precedence when they conflict.
Good uses for the global instruction:
- "Use bullet points, not paragraphs."
- "Prefer clinical terminology over layperson language."
- "Be concise--no more than two sentences per finding unless the finding is abnormal."
- "When the patient says 'right' and 'left,' always convert to OD and OS."
Do not duplicate field-specific guidance here. If you want a specific anterior segment format, put it on the Anterior Segment field, not in the global instruction.
Examples of effective instructions
Chief Complaint
Field: Chief Complaint (/background/patientReason)
"Format as a single sentence: '[Symptom/reason] x [duration].' If the patient gives multiple complaints, list the primary one first. Include laterality when stated."
Result: "Blurry vision at near OS x 6 months" instead of "Patient reports blurry vision, mainly in the left eye, has been going on for about six months."
Anterior Segment
Field: Anterior Segment (/examination/anteriorSegment)
"Grade pterygia as I/II/III/IV. Grade nuclear sclerosis as trace/1+/2+/3+/4+. Always note cortical and PSC status. For conjunctival findings, specify location (nasal, temporal, inferior, superior)."
This gives the AI a specific grading system to apply rather than free-form descriptions.
Posterior Segment
Field: Posterior Segment (/examination/posteriorSegment)
"For drusen, specify: type (hard/soft), location (macular/extramacular/peripapillary), and density (few/scattered/numerous). Always report C/D ratio as a decimal (e.g., 0.3, not 'small'). Note the presence or absence of a posterior vitreous detachment."
Assessment and Plan
Field: Plan (/assessment/plan)
"Always include follow-up timing in the plan. Format as numbered list. Each diagnosis should have its own plan item with: (1) current management, (2) next steps, (3) return interval. End with patient education if discussed."
Field: Follow Up (/assessment/followUp)
"Express follow-up as a number and unit, e.g., '6 months' or '2 weeks.' If the patient should return sooner for a specific concern, list both the routine and urgent return timelines."
Refraction
Field: Manifest Refraction (/examination/refraction/manifestRefraction)
"Always include the full Rx in minus cylinder form. Include vertex distance when over +/-4.00D sphere. Note the best corrected visual acuity at distance and near for each eye."
Tips for writing good instructions
Be specific. "Write better anterior segment notes" tells the AI nothing. "Grade nuclear sclerosis as trace/1+/2+/3+/4+ and always note lens status" gives it a clear target.
Use "always" and "never" for requirements. These signal absolute rules: "Always include laterality," "Never use the term 'unremarkable' -- instead list the specific normal findings."
Test with your next encounter. Instructions take effect immediately. After saving one, run your next patient encounter and check the output. Iterate quickly.
Keep instructions under two or three sentences. The instruction is appended to the field's schema description. Overly long instructions can dilute the AI's attention on the clinical content.
One instruction per field. You cannot stack multiple instructions on the same field. Combine your preferences into a single, clear directive.
Common mistakes
Too vague. "Be more detailed" is not actionable. The AI does not know which details you want. Say what you want included.
Conflicting instructions. If your global instruction says "Be concise, two sentences max" and a field instruction says "Include a full paragraph describing the appearance, location, extent, and grade," the field instruction wins--but the AI may produce inconsistent results across sections. Make sure global and field-level instructions complement each other.
Instructing a field that does not exist. The autocomplete in Settings shows only the ~40 fields that Doctora can actually instruct. If you type a custom path manually, the instruction may not resolve to any schema field. Stick to the suggestions.
Forgetting the global instruction. If you want all sections to use bullet points, set it once in the global instruction rather than repeating it across 15 fields. The global instruction exists precisely for cross-cutting preferences.
Over-constraining. If you require a specific format for every field, the AI spends more effort on formatting compliance and less on extracting clinical content. Instruct the fields that matter most to your workflow and let the defaults handle the rest.