Turning de-identified session shorthand into finished progress notes
The note backlog is the single most common reason therapists try AI — writing help and note summarization top the APA's survey of actual use. Because consumer chatbots can't touch PHI, the workflow that works is jotting a few de-identified shorthand lines right after session, letting AI structure them into a DAP or SOAP note, then reviewing before the note enters the EHR.
You are a clinical documentation assistant for a licensed {{license_type}}. Convert my de-identified session shorthand into a {{note_format}} progress note ready to paste into an EHR. Session shorthand (contains no identifying information): {{session_shorthand}} Rules: - Use only what is in my shorthand. Never invent symptoms, quotes, risk statements, interventions, or client history. If a standard element of a {{note_format}} note is missing (mental status observations, risk assessment, response to intervention, plan for next session), insert [CLINICIAN: add] rather than filling it in. - Name interventions in specific clinical language a payer recognizes (cognitive restructuring, behavioral activation, exposure with response prevention, motivational interviewing) — but only where my shorthand actually supports that intervention. - Write in third person, past tense, objective and behavioral ("Client reported...", "Client demonstrated...") with no diagnostic speculation beyond what I stated. - Keep it under 250 words. Do not add a diagnosis unless I provided one. - After the note, list anything an insurance reviewer would expect for this session type that is still missing.
Fill in your details and the prompt updates live — then copy.
Data: Client attended a 45-minute telehealth session. Reviewed thought-record homework, completed four of seven days. Client reported anxious mood related to an upcoming job review. Affect was anxious initially, brightening by end of session. Assessment: Client is engaging with cognitive restructuring and demonstrated ability to challenge catastrophic predictions with support. Presentation remains consistent with the existing GAD diagnosis. [CLINICIAN: add risk assessment] Plan: Continue weekly sessions. Homework assigned: daily thought record plus one grounding practice per day. Next session in one week. Still missing for a reviewer: risk screen, progress toward a specific treatment-plan objective.
The full workflow
- Write 3-6 lines of shorthand immediately after session, with every identifier stripped
- Run the prompt with your preferred note format
- Verify every clinical statement against your memory of the session — delete anything you did not actually observe or do
- Resolve every [CLINICIAN: add] flag yourself, especially risk
- Paste the finished note into your EHR and sign it
Watch out for
HIPAA: consumer ChatGPT, Claude, and Gemini accounts come with no Business Associate Agreement. Names, initials plus dates, or details that make a client recognizable are PHI — de-identify fully or use a BAA-signed tool (Mentalyc, Upheal, Blueprint, or your EHR's scribe).
You sign the note, you own it. AI drafts insert plausible clinical boilerplate — an intervention in the record that did not happen in the room is a false record.
Never let AI write the risk assessment. Suicide and homicide risk documentation must come from your direct assessment, in your words.
Where this comes from
Every use case on this site is grounded in real reports from working therapists — not invented by us.