Kipu Note Assistant helps clinicians generate structured documentation from transcripts such as intake conversations, dictations, or therapy sessions. While the AI is powerful, the default field labels like "HPI" or "Intervention" present in the template often don’t provide enough context for the AI to consistently produce high-quality notes aligned with your expectations.
That’s where Custom AI Prompts come in. By adding clear, specific instructions to each form field, you give AI the context it needs to generate documentation that is complete, compliant, and clinically useful.
💡 Tip: The more you invest in writing strong prompts, the greater your return in efficiency and accuracy. Well-crafted prompts reduce editing time, improve documentation consistency, and help the AI capture the right details the first time.
Why Prompts Matter
Without specific guidance, AI may generate vague or incomplete documentation. Adding a clear prompt gives the AI the structure it needs to produce clinically useful output.
❌ Without Prompt
Input: When the Kipu Note Assistant attempts to fill out a structured form using a transcript, it looks at the Name and Label for each form field and attempts to create appropriate documentation.
Output: Without specific guidance, AI may generate vague or incomplete documentation.
✅ With Prompt
Input: To help the AI better understand each form field’s purpose, we've included a Custom AI Prompt section in every text element. This allows you to provide specific guidance and context so the AI can generate more accurate, relevant documentation.
Output: The prompt turns raw transcript data into clear, useful clinical documentation.
The difference is not the transcript — it’s the prompt. Clear prompts serve as guardrails for the AI, helping it generate documentation that meets both clinical and organizational standards.
Where to Add Custom AI Prompts
Each text field within your evaluation forms includes a Custom AI Prompt field. There, you can define the prompt the AI will use when interpreting and documenting based on the transcript. Check out the Customize Evaluation Templates section of this article for more instructions.
Note: The AI relies heavily on the prompt field, but will default to the Label and Name of the field if no custom prompt is available. Sometimes, you may find success by simply providing better definition within these fields. Labels or Names like "HPI" or "Intervention" alone do not provide enough context for robust documentation.
The 6-Step Prompt Builder Framework
Use this framework to build effective Custom AI Prompts. Each step helps clarify what the AI should generate — and how to guide it with strong clinical instructions.
| Step | What to Do | Example |
|---|---|---|
| 1. Define the Purpose | Clarify what the field should capture in the clinical note. |
HPI: Patient’s current condition SIRP: Interventions and responses |
| 2. Break Into Components | Identify the key data points or clinical elements to include. | Onset, context, characteristics, aggravating/relieving factors, associated symptoms, impact |
| 3. Phrase as Prompts | Convert each component into a clear, directive question. | When did the problem begin and how has it changed? What makes symptoms worse or better? |
| 4. Add Documentation Tips | Include best practices for tone, format, and clinical language. | Best Practice: Be specific. Use measurable and observable language. Include denials when clinically relevant (e.g., denies suicidal ideation). Capture patient voice using quotes. |
| 5. Provide a Checklist | Create a one-line reference summarizing the expected content. | Chief Concern | Onset & Course | Context | Characteristics | Aggravating/Relieving | Associated Symptoms | Impact on Functioning |
| 6. Add a Short Example (Optional) | Give a realistic sample of a high-quality output. | Patient reports panic attacks for 3 months, worsened by stress, relieved by deep breathing; associated with insomnia and irritability; impairing work performance. |
Example Prompts by Field
| Form Field | Custom AI Instructions | Additional Comments |
|---|---|---|
| HPI | The HPI is the narrative clinical account of the patient’s current condition. Attempt to create a narrative using: Chief Concern, Onset & Course, Context, Characteristics, Aggravating/Relieving Factors, Associated Symptoms, and Impact on Functioning. Reflect emotional tone. | Use quotes. Capture emotions. Be chronological and clinically focused. |
| Situation/Symptoms | Create documentation from a behavioral health perspective. Include presenting problems, mood, affect, self-reported symptoms, and significant events. Use measurable criteria and quotes. | Be specific. Avoid vague phrases like "had a hard week." |
| Intervention | Describe therapeutic methods, skills introduced, delivery method (discussion, modeling, role-play), and link to treatment goals. Use clinical, action-oriented language. | Avoid generic descriptors like "counseled patient." |
| Response | Document verbal and nonverbal client responses, including quotes, tone, insight, and engagement level. Indicate progress toward treatment goals. | Write as a narrative. Use both patient voice and clinician observations. |
| Plan | Summarize progress and outline the plan for future interventions and next session goals. Include measurable objectives. | Stay future-focused. Maintain alignment with treatment plan. |
Tuning and Refining Prompts
AI prompting is an iterative process. You may need to adjust the prompt to achieve the best results.
Common Adjustments
- Too long? Add Limit to 3–4 sentences.
- Missing tone? Add Reflect patient’s emotional state.
- Unstructured? Add Format as bullet points.
Advanced Prompt Techniques
Add more control and precision with targeted phrasing:
- Format: Summarize in SOAP format.
- Length: Limit to 100 words.
- Tone: Use professional, clinical language.
- Perspective: Write in the voice of a behavioral health provider.
Tip: Treat AI output like a draft. Review, revise, and refine to get optimal results.
Prompt Refinement Example
Watch how refining your prompt transforms vague AI responses into structured, clinically useful documentation.
🗣 Transcript:"I've been having panic attacks for 3 months. Stress at work makes it worse, but breathing helps a little."
❌ Weak Prompt: "Write HPI."
AI Output: “Patient not feeling well.”
✅ Strong Prompt:"Include onset, context, triggers, and impact on functioning."
AI Output: “Patient reports panic attacks for 3 months, worsened by work stress, relieved by deep breathing; impairing functioning.”
✨ Tuned Prompt:"Limit to 3 sentences, reflect emotional tone."
AI Output: “Patient reports panic attacks for 3 months, worsened by stress, partially relieved by deep breathing, impairing daily life.”
Clinical Guardrails
AI is a support tool — not a replacement for clinician judgment. Always:
- Review output for clinical accuracy
- Document relevant denials
- Use quotes where possible
- Describe what you can observe
Important: Clinicians are responsible for the final documentation. Always verify and adjust AI-generated content before saving.
Comments
0 comments
Article is closed for comments.