AI medical scribing is a $397M market growing to $2.96B by 2033. Most clinicians don't need the enterprise tier.
Freed AI runs $59-119/month. Microsoft Dragon Ambient eXperience (DAX) runs $400+/month. DeepScribe is $400-700. Abridge is $400+ depending on integration. The category is growing 25%+ per year because the underlying problem — clinicians spending 2 hours on EHR for every 1 hour with a patient — is structural and worsening.
For solo physicians, small specialty practices, and NPs running independent clinics, the math at $400/month per provider is real. If you see 60-100 patients a week and can shave 30 seconds per encounter on documentation, the tool pays for itself. If you see 20 patients a week at lower-acuity, the math is tighter.
This post breaks down what those AI medical scribes are actually doing, why the value lives in the documentation prompt patterns (not the SaaS UI), and what a $49 lifetime pack of those patterns looks like — with appropriate disclaimers.
Critical disclaimer up front
This pack and this blog post describe a documentation assistance tool, NOT a clinical decision-making tool. Every output produced by these prompts must be reviewed and edited by a licensed clinician before being entered into a patient record, sent to a patient, used for billing, or used for any clinical decision.
Claude will sometimes hallucinate codes, drug names, or dosages. Verify every clinical detail against authoritative sources before relying on it.
The pack is for documentation assistance. You as the clinician remain responsible for every clinical decision and every record entry.
What Freed AI, DAX, and DeepScribe actually do
Strip the marketing and the AI medical scribe category has converged:
- Encounter capture. Listen to the patient encounter (with consent) and produce a structured SOAP note or H&P.
- Coding suggestions. Suggest ICD-10 and CPT/E&M codes based on the documented note.
- Note customization. Adapt to the clinician's specialty and documentation style over time.
- EHR integration. Push the completed note into Epic, Cerner, athenahealth, or another EHR.
- Patient summaries. Generate plain-language summaries for patient portals.
- Quality measure tracking. Surface MIPS/MACRA elements that need attention.
What's in the SaaS that's NOT in the prompt pack: the encounter audio capture, the EHR push, the BAA infrastructure, and the customer support. What IS in the pack: the documentation patterns that turn raw input into a clean clinical note.
For clinicians who type or dictate their notes (rather than relying on ambient capture), the pack handles the part that takes time — turning fragments into structured clinical documentation.
What makes a clinical AI prompt actually work?
The Clinical / Medical Documentation Pack uses the same seven-pattern framework as the other packs, with one critical addition: safety flagging.
For mental-health intake notes, the prompt explicitly flags any incomplete safety assessment. If the source notes mention suicidal or homicidal ideation without a complete safety assessment, the output marks [SAFETY ASSESSMENT INCOMPLETE — clinician must address before note is finalized]. The same pattern applies to documentation gaps that affect billing compliance, audit risk, or clinical correctness.
This isn't a feature most generic AI tools have because their training emphasizes producing complete output. Clinical work needs the opposite — explicit flagging of incompleteness is safer than confident filler.
The other six patterns (detailed in the seven-patterns post) apply standard: role anchoring ("you are a clinician converting raw encounter notes into a SOAP note"), context fences ("you have the encounter notes, you do not have prior chart history"), output contract ("SOAP format with specific section requirements"), failure mode declaration ("do not fabricate codes; flag insufficient documentation instead"), reasoning scaffolding (plan-first for complex differentials, answer-first for SOAP structure), iteration anchors ([VISIT_TYPE: complete H&P / problem-focused / follow-up], [SPECIALTY: family medicine / dermatology / mental health]).
Three example prompts from the Clinical Pack
Example 1: SOAP note from raw encounter notes
The prompt takes raw notes (verb fragments, observations, findings) and produces a SOAP-format clinical note with proper subjective, objective, assessment, and plan sections.
The critical structural element: documentation-gap flagging. If a complete review of systems is required for the visit type but not in the source notes, the output flags [DOCUMENTATION GAP: ROS not noted] rather than fabricating. The clinician fills the gap or down-codes the visit appropriately.
This is the part the SaaS tools and even most prompt engineering courses skip. Confident filler is the worst possible output for clinical documentation because it survives chart review by looking complete and gets the clinician audited or sued.
Example 2: ICD-10 and CPT code suggestion with rationale
The prompt suggests codes based on the documented note and provides the specific note language supporting each code. Critically, the prompt does NOT code conditions that aren't documented. "Patient looks like they might have depression" doesn't produce an F33 code. Documented "PHQ-9 score 18, persistent low mood, anhedonia x 4 weeks" does.
The prompt also flags codes that are commonly downcoded on audit or where the documentation may be insufficient — internal flags so the clinician can decide whether to upgrade the documentation, downcode the bill, or both.
Mandatory disclaimer: tax law, ICD-10, and CPT change constantly. AI is known to use outdated versions. Every code suggestion must be verified against the current ICD-10-CM and CPT manuals before submission.
Example 3: Plain-language diagnosis explanation at 6th-grade reading level
The prompt converts a clinical diagnosis into a written explanation a patient can read and understand. Sentence length under 15 words. Avoid words longer than 3 syllables unless they're essential (and then define them inline). Reading level: 6th grade.
Why this matters: most patient-facing AI tools produce explanations at a 10-14th grade reading level. The average US adult reads at 7th-8th grade. Patient comprehension of their own diagnosis directly correlates with treatment adherence. Plain language is medical efficacy, not just accessibility.
What the pack contains
The Clinical / Medical Documentation + Claude Code Prompt Pack ships v1.0 with 8 starter prompts (2 per role) at full quality, expanding to all 50 prompts in v1.1 within 5 weeks of launch.
Clinical Documentation: SOAP notes from raw encounter notes, ICD-10 / CPT code suggestion with rationale, differential diagnosis bullet from chief complaint, treatment plan with patient-friendly explanation, medication reconciliation summary, discharge summary draft, admission H&P, comprehensive vs problem-focused visit calibration, telemedicine documentation, after-hours encounter documentation, controlled-substance prescribing documentation, MDM justification for E/M level, time-based billing documentation, chronic-care management, transitional-care management.
Specialty-Specific: mental health intake assessment (with explicit safety flagging), physical therapy progress notes, dermatology lesion descriptions, pediatric well-child visits, geriatric medication review, OB prenatal visits, endocrine medication adjustments, cardiology consult, GI procedure, sports medicine return-to-play, pre-operative clearance, occupational therapy progress, speech-language pathology.
Patient Communication: plain-language diagnosis explanations (6th-grade reading level), lab-result follow-up emails, appointment-prep instructions, denial-of-care explanations, referral-handoff letters, medication-change explanations, post-op instructions, second-opinion handling, patient-portal message replies (10 common scenarios), no-show follow-up, family-member-as-proxy communication, end-of-life care discussions.
Practice Operations: prior-authorization letter structure, denied-claim appeal letters, peer-to-peer call talking points, ICD-10 lookup chain, MIPS quality measure narratives, FMLA paperwork, disability evaluation, advance-directive documentation, refill-request workflow, controlled-substance verification.
Full pack at clskillshub.com/pack/medical.
Why $49 lifetime vs $59-700/month
The SaaS scribes charge monthly because they bundle: ambient audio capture, EHR integration, customer support, BAA infrastructure, and venture-investor returns. For clinicians whose practice depends on ambient capture during encounters, those tools earn their cost.
The pack ships the documentation prompt patterns. You bring: your own Claude or Anthropic account (preferably an enterprise tier with a BAA if you're handling PHI), your own audio-to-text if you want it (Apple Dictation, Otter, or a typist), and your own EHR. The trade is convenience for ownership.
For clinicians who type or dictate their notes today, the pack is the upgrade that makes typing 3x faster. For clinicians who depend on ambient capture, the SaaS is still the right tool.
A note on HIPAA
The pack itself is just text — Markdown files containing prompt patterns. HIPAA compliance depends entirely on how you USE the prompts.
If you copy PHI into a Claude session, you need a Business Associate Agreement with Anthropic. Anthropic offers HIPAA-compliant enterprise tiers; consumer Claude.ai does not have a BAA.
If you de-identify before pasting (replace patient details with generic placeholders), HIPAA generally doesn't apply because there's no PHI in the flow. Many clinicians use this approach for general documentation patterns, then re-personalize the output in their EHR.
The pack documentation includes specific guidance on which prompts can be used with de-identified data and which require a BAA. Read the disclaimers in the pack before using with PHI.
Who should buy this
Solo physicians, small specialty practices (PT, OT, mental health, dermatology, family practice), NPs running independent clinics, and residents/fellows learning charting who:
- Type or dictate their own clinical notes today
- Want structured prompts that produce complete documentation consistently
- Are willing to verify codes, drugs, and clinical content against authoritative sources
- Are on a Claude tier with BAA if using with PHI (or de-identify before use)
- Don't need full ambient encounter capture
NOT for: large hospital systems with enterprise EHR contracts, clinicians who require full ambient encounter capture during patient visits, or anyone who needs hand-holding through HIPAA compliance. The pack assumes a level of professional judgment and HIPAA awareness.
How to use the pack
1. Copy-paste into Claude. Run on a BAA-covered tier if handling PHI. De-identify first if using a consumer tier.
2. Install as Claude Code skills. Drop skills-install/medical/ into ~/.claude/skills/, restart Claude Code, type @medical in any session. Verify your account configuration before using with PHI.
3. Searchable reference. Ctrl+F for "SOAP", "ICD", "prior auth", "discharge".
Related reading
- The 7 Patterns Behind High-Performance Claude Prompts — the framework every prompt uses
- Harvey AI Alternative for Lawyers — adjacent vertical with similar structural patterns
- Pricing page — all 7 industry packs
- Clinical Pack on clskillshub.com
FAQ
How does this compare to Freed AI or DAX? Freed AI and DAX are ambient audio scribes that listen to the patient encounter and produce a SOAP note. The Clinical Pack is documentation prompt patterns — you provide the text input (typed, dictated, or pasted from another tool), the prompts produce the structured note. Different workflow. If you need ambient capture, stay on Freed or DAX. If you type your notes today, the pack is the upgrade that makes typing fast.
Is it HIPAA-compliant? The pack is just text. Your USE of the pack must be HIPAA-compliant — that means running on a BAA-covered Claude tier when handling PHI, or de-identifying before use. The pack includes per-prompt guidance.
Will Claude hallucinate codes or drugs? Yes, occasionally. Every code, drug, and dosage must be verified against authoritative sources before relying on it. The pack's prompts are designed to flag uncertainty rather than confidently fabricate.
Can this replace my EHR? No. The pack helps you produce clinical documentation; you still enter it into your EHR. The pack reduces the time spent on the documentation; the EHR is still the system of record.
Lifetime updates? Yes.
Refund policy? Digital product, all sales final. If something genuinely does not land for you, reply to your purchase email and I will add you to the full Skills Library (lifetime access) as a goodwill gesture.