Writing Effective Form Labels for AI Scribe
AI Scribe generates clinical documentation by analyzing your session transcripts and filling in form fields based on the field labels you provide. The label is the primary signal Scribe uses to understand what content belongs in each field — a vague label produces vague output.
This guide covers how to write field labels that produce accurate, consistent charting notes.
Looking for Custom Instructions guidance?
Field labels are the first thing to optimize, but for tone, clinical logic, and cross-field rules, see the AI Scribe Configuration Guide.
How Labels Work
Every field label is an instruction to the AI. Scribe reads your label to decide:
- What type of information belongs here
- Whether to pull from this session's transcript or preserve existing content
- How specific or broad to be in what it captures
If a field is consistently mis-filled or left blank, the label is almost always the first place to look — before adding Custom Instructions.
Core Principles
1. Be specific and direct
Vague labels give Scribe too much room to interpret. Specific labels constrain the output to what you actually need.
| Instead of | Use |
|---|---|
| Notes | Subjective assessment — patient's reported concerns, symptoms, and experiences from today's session |
| Assessment | Clinical observations of client behavior, mood, and cognition explicitly noted by provider during session |
| Plan | Intervention plan — specific actions, referrals, or recommendations discussed this session |
| History | Medical history — conditions, diagnoses, and relevant background as reported by the patient |
2. Include terminology preferences in the label
If you want Scribe to use specific language — person-first language, your organization's preferred pronouns, or clinical terminology — put it in the label.
✅
Client assessment — refer to the individual as 'client' and use they/them pronouns unless otherwise documented
3. Focus on what was explicitly reported or observed
Labels that ask for inferred or estimated information (e.g. "anxiety level") give Scribe permission to guess. Labels that ask for reported or observed information keep the output grounded in what actually happened.
| Instead of | Use |
|---|---|
| Client's anxiety level | Client's self-reported anxiety symptoms and their described severity, using the client's own language |
| Emotional state during session | Client's explicitly stated emotional experiences during session — do not infer emotional states from tone or speech patterns |
Labels and Placeholder Text Work Together
For fields that need a specific structure or format, the label should identify what the content is, and the placeholder text should show Scribe how to format it.
- Label: identifies the content (
Chief Complaint — document a direct patient quote, 1–2 sentences) - Placeholder: models the output format (e.g. a bullet template, a multi-part structure)
This is especially useful for fields with sub-sections, dated entries, or bullet-formatted output. The placeholder acts as a formatting template the AI follows when generating the field's content.
Note: Bullet and list formatting cannot be controlled through Custom Instructions — it must be configured in the placeholder.
Field-Specific Label Guidance
Session content documentation
Therapy progress notes:
Summarize key therapeutic interventions used today, client responses, and progress toward treatment goals. Focus on evidence-based techniques discussed.
Trauma processing documentation:
Document specific trauma-related topics addressed, client's expressed thoughts about the trauma, and any cognitive restructuring work completed during this session.
Homework and between-session activities:
Record any therapeutic homework assignments given, client's completion of previous assignments, and their reported experiences with prescribed activities.
Clinical assessment fields
Symptom documentation:
Document symptoms explicitly reported by client during today's session. Only include symptoms the client directly described or discussed.
Safety assessment:
Document any safety concerns, risk factors, or protective factors explicitly discussed or observed during this session.
Administrative and continuity fields
Treatment planning:
Record treatment plan modifications discussed during session and client's expressed agreement or concerns about proposed changes.
Trauma-Informed Considerations
Labels for trauma-informed practices should be explicit about language, inference limits, and client agency.
Avoiding inappropriate inference:
Client's explicitly stated emotional experiences during session — do not infer emotional states from tone or speech patterns
Strength-based framing:
Document client's expressed strengths and coping strategies discussed today. Focus on resilience factors and the client's own words about their progress.
Cultural sensitivity:
Document any cultural, spiritual, or identity factors the client mentioned as relevant to their experience or recovery process.
Sensitivity limits:
For Trauma History: provide a high-level summary only — do not include specific acts, locations, or perpetrator information
Common Mistakes to Avoid
| Problem | Fix |
|---|---|
| Label is too vague | Add the field's specific clinical purpose and scope |
| Label leads Scribe to infer | Add "as reported by" or "explicitly discussed" to anchor it |
| Field is consistently mis-filled | Rewrite the label before adding Custom Instructions |
| You want specific formatting | Use placeholder text, not the label |
Quality Improvement Tips
- Start simple. Begin with basic labels and refine based on output quality.
- Iterate from output. If a field's content is off, read the label as if you were the AI — does it clearly say what you want?
- Fix the label first. Before adding a Custom Instruction to correct a field's behavior, try rewriting the label. It's faster and creates less room for conflicts.
- Use placeholder text for format. If you need structured output, model it in the placeholder — not the label.