Prompt Chaining: Batch End-of-Day Documentation System

What This Builds

Instead of writing progress notes one-by-one through your afternoon, this system lets you jot quick bullet points during or right after each patient session, then run ALL your notes through AI at the end of the day in one batch. What currently takes 90–120 minutes of stop-start documentation becomes 30–40 minutes of focused reviewing.

The setup involves building a master documentation prompt that processes multiple patients in a single AI request. You batch your session data, submit once, get all your drafts back, then review and paste into your EMR.

Prerequisites

• Comfortable using ChatGPT or Claude for individual notes (see Level 3 guide)
• ChatGPT Plus ($20/mo) or Claude Pro ($20/mo) — for longer context windows needed for batch processing
• A way to capture quick notes during sessions (phone notes app, pocket notepad, or tablet)
• About 1 hour to set up and test the system

The Concept

Think of this like prepping a batch of materials at a manufacturing station instead of one at a time. A factory worker who needs to cut 14 pieces of wood cuts all 14 measurements first, then runs them through the saw in sequence — much faster than measuring and cutting each piece individually.

You're doing the same: capture session data throughout the day (takes 30 seconds per patient), then process all 14 patients through the AI saw in one efficient batch.

Build It Step by Step

Part 1: Set Up Your Session Capture System

Option A: Phone Notes App Create a note template on your phone (iPhone Notes or Google Keep) that you fill in during or immediately after each session:

PATIENT: [initials or number only]
DX: [diagnosis]
S: [patient complaint/report — 1 sentence]
O: [treatments, exercises, measurements — bullets]
P: [next session plan — 1 sentence]

Set a shortcut or text replacement: type "soapt" → expands to the template.

Option B: Pocket Notepad Old-school but effective — a small notepad in your coat pocket. Write shorthand during documentation-heavy transitions. Convert at end of day.

Option C: Voice Memo After each patient exits the room, speak a 20-second summary into your phone's voice recorder. Transcribe at end of day (or use a voice-to-text app).

Pick whichever method feels most natural. The goal is capturing raw data in real time so you're not reconstructing it from memory 4 hours later.

Part 2: Build the Batch Documentation Prompt

After setting up your capture system, create this master prompt that processes multiple patients at once. Save it in a text file on your desktop:

I am a Physical Therapist in outpatient orthopedics. I need you to write SOAP progress notes for each of the following patients. Use professional PT clinical language. Keep each note under 200 words. Do not include patient names — I will add those in my EMR.

FORMAT FOR EACH NOTE:
Patient [number]
S: [subjective — 2 sentences]
O: [objective — 2-3 sentences including specific measurements and treatments]
A: [assessment — 1-2 sentences: progress and clinical impression]
P: [plan — 1-2 sentences for next visit]

---

HERE ARE TODAY'S PATIENTS:

Patient 1:
DX: [diagnosis]
Session data: [your raw bullet points from capture system]

Patient 2:
DX: [diagnosis]
Session data: [your raw bullet points]

Patient 3:
[continue for all patients]

---

Write all notes in order. Use clinical terminology appropriate for outpatient PT documentation.

Part 3: Test and Refine

Run a test with 3-5 patients before relying on this for a full day. Evaluate:

• Are the notes clinically appropriate? (Check terminology and structure)
• Are they the right length? (If too long, add "Keep each note under 150 words")
• Does the AI maintain consistent format across all patients?
• Is the output easy to copy-paste note by note into your EMR?

Adjust the prompt based on your results. Common refinements:

• Add: "Always include specific measurements in the objective section" if measurements keep getting omitted
• Add: "For Medicare patients, include functional progress toward goals in the Assessment section"
• Add: "Do not use the phrase 'the patient' — use 'patient' without the article"

Real Example: End-of-Day Batch Run

Setup: Tuesday clinic day. 13 patients seen. 5:45pm. You've captured session notes in your phone all day.

Your phone notes look like:

Pt 1: TKR 6wk. S: 3/10 pain, better than last week. O: bike 20min, SLR 3x15 5lb, flex 118°. P: progress to step training
Pt 2: rotator cuff repair 10wk. S: 5/10 overhead, no change. O: pulley exercises, ER 45° (up from 38°), scapular stabilization. P: continue same
Pt 3: lumbar disc. S: 4/10 pain, walked dog yesterday without issue. O: core stabilization 3x15, walking 20min, flexion 60°. P: progress to uneven terrain walking
...and so on for 13 patients

What you do: Open ChatGPT Plus, paste your master batch prompt with all 13 patient data entries, submit.

What you get: 13 complete SOAP notes in 20-30 seconds.

What you do next: Read through all 13 notes (5-10 minutes), correct any errors, paste into WebPT one at a time.

Time saved:

• Traditional: 13 notes × 10 min each = 130 minutes (2+ hours)
• Batch system: 13 notes captured during day (6 min) + batch prompt (2 min) + review/edit (15 min) + EMR entry (20 min) = ~45 minutes

Net savings: ~85 minutes on a 13-patient day.

What to Do When It Breaks

• AI cuts off before finishing all patients → Your prompt was too long for the context window. Break into two batches (first 7 patients, then 6). ChatGPT Plus and Claude Pro have longer contexts than free tiers.

• Notes are too generic or vague → Your raw session data is too sparse. Add a line to your prompt: "Use specific measurements and treatments from the session data provided — do not add generic language."

• Inconsistent format across patients → Add an explicit format example at the top of your prompt showing exactly what a correct note looks like.

• AI invented clinical details you didn't provide → This is a known risk — always review every note before EMR entry. If you see measurements you didn't record, delete them.

Variations

• Simpler version: Process 5 patients at a time instead of a full day's caseload — easier to review in chunks.

• Extended version: Add a second prompt pass where you submit all notes back and ask the AI to "flag any notes that are missing specific measurements or that may not adequately support the CPT codes for a 15-minute timed unit" — adds a billing compliance check layer.

What to Do Next

• This week: Test the system on a 5-patient batch before using it for a full day
• This month: Refine the capture method that works best for your flow — phone notes vs. voice memos vs. notepad
• Advanced: Add a second prompt that processes your batch output and generates a summary of any patients who need attention (no progress noted, pain escalating, goal not on track) — daily clinical caseload review in one shot

Advanced guide for physical therapist professionals. All AI-generated clinical documentation must be reviewed for accuracy before entering into your EMR. De-identify all patient data before submitting to AI tools without a HIPAA BAA.