Prompt Chain: Systematic Audio Blank Resolution Workflow

For Medical Transcriptionists

Tools: ChatGPT (free or Plus) | Time to build: 1 hour | Difficulty: Intermediate-Advanced Prerequisites: Comfortable using ChatGPT for terminology research — see Level 3 guide: "Deep Specialty Research for Unfamiliar Procedures"

What This Builds

A three-step prompt chain that turns the chaotic process of resolving audio blanks into a systematic workflow: identify the most likely terms, research each option in clinical context, and then draft the physician clarification message — all in one focused ChatGPT session. What currently takes 15-20 minutes of fragmented effort (replaying audio, Googling, writing emails) becomes a structured 5-8 minute process.

Prerequisites

• ChatGPT account — free tier works for this workflow
• Access to the dictation audio and surrounding transcribed text
• Your blank's clinical context: specialty, report type, surrounding sentences

The Concept

A prompt chain is like an assembly line — each prompt takes the output of the previous step and builds on it. Instead of asking one vague question and getting a generic answer, you ask three focused questions in sequence, where each answer informs the next. For blank resolution, the chain is: Identify → Research → Communicate. By the time you reach the physician communication step, you've already done the clinical reasoning and can draft a much more informed message.

Build It Step by Step

Part 1: Set Up Your Blank Resolution Session

1. Open ChatGPT (free at {{tool:ChatGPT.url}} — no paid plan required)
2. Start a new chat
3. Orient ChatGPT with your context — this is the setup prompt you run once per blank:

Orientation prompt:

I'm a medical transcriptionist working on a [specialty] [report type]. I have an audio blank — a section where the physician's dictation is unclear. I need your help resolving it. I will walk you through a 3-step process.

Context for this blank:
- Specialty: [e.g., cardiology, orthopedics, radiology]
- Report type: [e.g., operative note, discharge summary, clinic note]
- Surrounding text before the blank: "[paste the sentence(s) before the unclear section]"
- Surrounding text after the blank: "[paste the sentence(s) after the unclear section]"
- The audio sounds like: "[your best phonetic approximation — even rough]"
- Patient context (NO names or identifiers): [e.g., "elderly male with history of CHF undergoing cardiac catheterization"]

Ready to start Step 1?

What you should see: ChatGPT acknowledges the context and is ready for Step 1.

Part 2: Step 1 — Identify the Most Likely Terms

Now run Step 1 of the chain:

Step 1 prompt:

Step 1: Candidate Identification

Based on the clinical context I gave you, what are the 3-5 most likely medical terms, medications, or procedures that could fill this blank? For each candidate:
- List the term with correct spelling
- Explain why it fits the clinical context
- Rate the likelihood: High / Medium / Low

Rank them by likelihood.

What you should see: A ranked list of candidates with clinical reasoning — not a generic list of medical terms, but specific candidates that fit the exact context you provided.

Part 3: Step 2 — Clinical Research for Each Candidate

Take the top 2-3 candidates from Step 1 and research them in context:

Step 2 prompt:

Step 2: Clinical Verification

For the top 2-3 candidates from Step 1, verify each one:
1. What is the standard dosage or usage for this term in a [specialty] context? (if it's a medication)
2. Would this term normally appear in the position it's in (before "[text before]" and after "[text after]")?
3. Is there any reason this term would be clinically inconsistent with the patient context I described?

Based on this analysis, which single candidate do you recommend as most likely correct?

What you should see: A clinical verification of each candidate with a clear recommendation. At this point, you either have your answer (high confidence) or you know exactly what to ask the physician.

Part 4: Step 3 — Draft the Physician Communication

Use Step 2's output to draft your physician message:

Step 3 prompt:

Step 3: Physician Communication

Based on our analysis, draft a brief, professional email to the physician requesting clarification on this blank.

Include:
- The specific location in the report (section name + approximate position)
- Your best guess for the term (from Step 2) so the physician can confirm or correct quickly
- A request for clarification phrased politely and specifically
- Keep it under 4 sentences total

The physician's name is Dr. [last name]. The report is a [report type] from [date or description].

What you should see: A ready-to-send professional email that includes your best-guess term — making it easy for the physician to respond with a simple "Yes, that's correct" or "No, it's [correction]" rather than having to re-listen to the audio themselves.

Real Example: Cardiac Medication Blank

Setup context:

• Specialty: Cardiology
• Report type: Discharge summary
• Before blank: "Patient was continued on her home medications including lisinopril 10mg daily and"
• After blank: "Patient was instructed to take this twice daily with food"
• Audio sounds like: "low-oh-vee forty" or possibly "lah-vah-stah-teen"
• Patient context: Elderly female with hypertension and hyperlipidemia

Step 1 output: Top candidates — Lovastatin 40mg (High), Atorvastatin 40mg (High), Rosuvastatin 40mg (Medium) — all statins that would fit the hyperlipidemia context and "twice daily" doesn't match most statins (which are once daily), so this narrows it

Step 2 output: Most statins are once daily; Lovastatin is twice daily — this is significant. The audio approximation "low-oh-vee forty" most closely matches "Lovastatin 40mg." Recommendation: Lovastatin 40mg with a flag that twice-daily dosing is consistent.

Step 3 output: "Dear Dr. Chen, I'm transcribing the discharge summary from [date] and have a blank in the medications section. After 'lisinopril 10mg daily,' the audio is unclear — my best interpretation is 'Lovastatin 40mg twice daily.' Could you confirm or correct this? The report is otherwise complete and ready for your review."

Time: 6 minutes total vs. 20 minutes of fragmented manual research + email drafting.

What to Do When It Breaks

• Step 1 gives too many candidates → Add more clinical context to your orientation prompt — the more specific the surrounding text, the fewer candidates survive
• Step 2 recommendation is uncertain → Tell ChatGPT "The audio sounds more like [revised phonetic] — does this change your recommendation?" to iterate
• Step 3 email is too long → Add: "Make it shorter — maximum 3 sentences"
• ChatGPT doesn't know the specific drug or device → Paste the Step 1 candidate list into a specialty-specific reference (drugs.com, Stedman's) for final verification

Variations

• Simpler version: Run just Step 1 as a quick terminology lookup without the full chain — when you're confident after Step 1, skip Steps 2 and 3
• Extended version: Add a Step 4 — after the physician responds, paste their correction into a final prompt to add it to your terminology database automatically

What to Do Next

• This week: Run the full 3-step chain on your next 5 audio blanks. Time it vs. your current method.
• This month: After using it regularly, identify which Step is generating the most value and double down on that part of the workflow
• Advanced: Combine this chain with the Custom GPT draft reviewer (Level 4 guide) for a complete AI-assisted QA workflow

Advanced guide for medical transcriptionist professionals. Free ChatGPT account is sufficient for this workflow. Tool interfaces may change.