Independent practices, group clinics, urgent care, home health, hospice, allied health, behavioral health, dental, and small specialty offices run on documentation, communication, and patient education. AI's leverage in healthcare is in the work that surrounds the visit — charting drafts, patient letters, prior auths, intake summaries, SOPs, and follow-ups. AI does not diagnose, does not treat, and does not replace clinical judgment. It drafts. The clinician signs.
Highest-leverage use cases
Where AI actually earns its keep.
7 concrete plays we’ve seen consistently work in healthcare & clinical. Time-saved estimates are conservative.
Visit note draft from dictation
Medium
Convert a clinician's voice memo or dictation into a structured SOAP/DAP note draft. Clinician reviews, edits, and signs every note before it enters the chart.
5-8 hrs / clinician
Patient education materials
Easy
Generate plain-language, reading-level-controlled handouts for specific conditions, post-visit instructions, or medication education. Reviewed by clinician before patient delivery.
2-4 hrs / week
Prior authorization & appeal letters
Medium
Template-based prior auth and denial-appeal letters built from chart context the clinician provides. Letter cites medical necessity in payer-friendly language.
3-6 hrs / week
Patient intake summarization
Easy
Turn long intake forms or pre-visit questionnaires into a structured pre-visit summary the clinician can scan in 60 seconds. De-identified inputs only when using non-BAA tools.
2-3 hrs / clinician
Referral & consult letters
Easy
Draft referral letters to specialists or consult-back letters from chart-validated facts. Clinician verifies and signs before sending.
2-3 hrs / week
Patient follow-up & recall communication
Easy
Post-visit check-ins, no-show recovery, annual recall, and overdue-labs outreach — written in your practice voice, no clinical advice in the message itself.
3-5 hrs / week
Clinical SOPs & front-desk scripts
Easy
Rooming workflows, sterilization SOPs, intake scripts, scheduling-triage scripts (administrative, not clinical). The work nurses end up writing on lunch breaks.
2-4 hrs / week
Sample prompts · ready to paste
Prompts that actually work.
Specific, role-tagged, with guardrails baked in. Drop into Claude, ChatGPT, or your AI tool of choice.
SOAP note draft from dictation
For: Clinician (MD / DO / NP / PA / RN / DPT / SLP)
You are drafting a SOAP note from a clinician's dictation. Output structure: Subjective, Objective, Assessment, Plan. Use only what the clinician dictated — do not infer findings, diagnoses, or plans that were not stated. Keep clinical language; do not soften clinical terms.
Dictation: [PASTE OR TRANSCRIBE]
Visit type: [NEW / FOLLOW-UP / ACUTE / WELLNESS]
Patient context (de-identified): [AGE RANGE, SEX, RELEVANT HX]
Do NOT add ICD/CPT codes. Do NOT recommend treatment beyond what the clinician stated. Flag any section where the dictation was unclear with [CLARIFY]. This is a DRAFT — clinician reviews and signs before the note enters the chart.
DAP note for behavioral health
For: Therapist / counselor / behavioral health clinician
Draft a DAP note (Data, Assessment, Plan) from session notes. Use clinical-but-humane language. Preserve clinician's clinical impressions exactly as stated; do not add diagnostic impressions of your own.
Session notes: [PASTE]
Modality: [CBT / DBT / EMDR / SUPPORTIVE / OTHER]
Session #: [N]
Do NOT add a diagnosis the clinician did not state. Do NOT generate risk assessments — clinician documents risk directly. Flag any ambiguity with [CLARIFY]. Output is a draft for clinician review and signature.
Patient education handout
For: RN / MA / clinician
Write a patient education handout on [CONDITION OR TOPIC] for a patient at a [READING LEVEL — default 6th grade] reading level. Structure: What it is, Why it matters for you, What you can do, When to call us, When to go to the ER. Use plain language, short sentences, no jargon (or define it on first use).
Patient context: [AGE RANGE, RELEVANT BACKGROUND — de-identified]
Clinician's specific instructions to include: [LIST]
Clinician's specific instructions to EXCLUDE: [LIST]
Do NOT recommend specific medications, doses, or treatment changes. Do NOT contradict the clinician's instructions. End with: 'Reviewed by [CLINICIAN NAME] on [DATE].' This handout is reviewed and approved by the clinician before any patient receives it.
Plus 13 more prompts in the full pack
The complete Healthcare & Clinical pack ships in our Company AI Day — including agent templates, compliance notes, and the full prompt library.
HIPAA & BAAs: free Claude and free ChatGPT do NOT have BAAs and must NEVER receive PHI. Claude Team/Enterprise and ChatGPT Enterprise offer BAAs — confirm yours is signed and on file before any PHI-adjacent use, and document which tools are approved in writing.
PHI minimization (Safe Harbor): paste only what's necessary. De-identify per HIPAA Safe Harbor's 18 identifiers (names, dates more specific than year, geographic units smaller than state, MRNs, etc.) when using any tool that doesn't have a BAA, and prefer de-identification even when one does.
AI is not a clinician: every AI output that touches a chart, a patient, a payer, or a clinical decision requires clinician review and sign-off. AI drafts. Clinician signs. Make this explicit in your written policy.
Patient-facing AI disclosure: states are moving fast on this. Utah and California have AI-disclosure rules touching healthcare communications, and more states are following. Check your state and update intake forms and patient communications accordingly — and re-check every 6 months.
Coding & billing risk: overcoding and undercoding are both billable as fraud. AI may suggest codes, but a credentialed coder or clinician verifies every code before submission. Do not let an AI assign ICD/CPT autonomously.
Hallucination risk for clinical references: general LLMs fabricate dosing, drug interactions, guideline citations, and policy numbers with confidence. Never trust clinical references from a general AI — verify against UpToDate, Lexicomp, FDA labeling, or the actual guideline document.
Telehealth and EMR integration: any AI tool that touches your telehealth platform or EMR must be vetted for BAA and data-flow before connection. 'It works' is not the same as 'it's compliant.'
Written AI policy & training: every practice needs a written AI use policy (approved tools, what data may be entered, sign-off requirements, incident reporting) and annual staff training. 'Everyone uses what they want' is not a policy — it's an audit finding.
Wins we’ve seen
Real outcomes.
A 4-provider family practice cut after-hours charting by ~40% by piloting a dictation-to-SOAP draft workflow with one clinician first, then rolling it out — every note still reviewed and signed by the clinician.
An outpatient PT clinic with 3 therapists built a patient education writer trained on their voice and standing exercise protocols; handout production dropped from 30 minutes to 5 minutes per topic, with the lead PT reviewing every handout before patient delivery.
A solo RN concierge practice built a HIPAA-safe prior auth letter generator on Claude Team and lifted first-pass approval rate by ~25% over a quarter — without any PHI leaving the BAA-covered environment.
Three ways forward
Make this real for your team.
Free
Run AI Day yourself
Free DIY playbook with the full Healthcare & Clinical pack — agenda, prompts, agent templates, the whole thing.
