Ready-to-adapt prompt templates for QA and compliance staff working on deviation documentation, audit preparation, SOP review, and stability impact assessment. Replace bracketed placeholders with your actual data, and never paste unredacted batch or patient-identifying data into a consumer-tier AI account.
Deviation Record Structuring
Use when: Converting raw investigation notes into a structured deviation record draft.
“Here are my raw investigation notes for deviation [ID], attached. Structure them into our standard deviation report format: description, immediate actions, investigation findings, root cause, impact assessment, CAPA. Do not add any root cause or conclusion that isn’t explicitly stated in my notes — if the notes don’t identify a clear root cause, state that the root cause is inconclusive rather than inferring one.”
Watch for: A confident-sounding root cause statement that isn’t actually traceable to your notes — see When Deviation Root Causes Get Invented for why this specific failure mode matters.
CAPA Drafting Tied to Root Cause
Use when: Drafting corrective and preventive actions once a root cause has been confirmed by the investigation.
“The confirmed root cause for deviation [ID] is: [state root cause]. Draft 2-3 corrective actions and 1-2 preventive actions that specifically address this root cause — not generic process improvements. Explain how each action connects back to the stated root cause.”
Watch for: CAPA language that sounds thorough but doesn’t actually map to the stated root cause — a common quality issue that AI can inherit if the prompt doesn’t force the connection explicitly.
GDP Audit Gap Analysis
Use when: Preparing for an upcoming GDP inspection by checking current SOPs against the applicable guideline.
“Here is our SOP for [topic], attached, along with the relevant sections of [applicable GDP guideline — e.g., EU GDP Guidelines]. Identify any areas where our SOP language appears inconsistent with the guideline, or where the guideline addresses something our SOP is silent on. Treat silence as a gap, not as implied compliance. Use only the attached guideline text — do not rely on general knowledge of GDP requirements.”
Watch for: The AI filling in what it “believes” the guideline requires instead of what the attached text actually says — regulatory text should always be supplied directly, not recalled from the model’s training data.
Mock Auditor Question Simulation
Use when: Rehearsing responses to likely auditor questions ahead of an inspection, especially around a known prior finding.
“Simulate a GDP auditor who previously cited us for [prior finding]. Ask me a series of follow-up questions probing whether our corrective actions actually resolved the issue, based on our current SOP [attached] and our last 3 internal audit results [attached]. Push back if my answers are vague or don’t match what’s actually in the documents.”
Watch for: Answers that sound polished but overstate what the documents actually support — treat this as rehearsal for accuracy, not for sounding confident.
Batch Record Completeness Check
Use when: Reviewing a batch record against its SOP template for completeness before release review.
“Here is our SOP template listing required fields and sign-off points, and this completed batch record, both attached. Check whether every required field and sign-off point is present and completed. Flag any missing entries, blank fields, or inconsistencies between sections — do not assume a blank field is acceptable just because it’s not flagged elsewhere.”
Watch for: The AI treating a blank or ambiguous field as acceptable by default — completeness checks should flag ambiguity, not resolve it silently.
Stability Impact Technical Summary
Use when: Organizing excursion data against stability study data to support (not make) a disposition decision.
“Here is the excursion data (duration, temperature readings) for shipment [ID] and our stability study summary for product [name], both attached. Summarize how the excursion’s cumulative time-out-of-refrigeration compares to the studied stability conditions. Describe the comparison only — do not state whether the product should be released, held, or rejected; that determination will be made by QA separately.”
Watch for: Any disposition-style language creeping into the summary — see When Disposition Boundaries Get Blurred for the specific pattern to guard against.
Supplier Questionnaire Cross-Check
Use when: Reviewing a supplier’s completed quality questionnaire against required supporting evidence.
“Here is the supplier’s completed quality questionnaire and their attached certifications, both attached, plus our internal record of complaints and deviations linked to this supplier over the past [timeframe]. Identify any questionnaire response that isn’t supported by an attached document, and any historical issue from our internal record that isn’t addressed in their response.”
Watch for: Accepting a self-reported certification claim without independently verifying it against the certifying body or a current certificate — the AI can flag inconsistency, but it cannot verify authenticity.
General Guidance for This Role
- Never let an AI-drafted root cause, disposition summary, or audit response go directly into a record or in front of an auditor without a documented human review step.
- Always supply the actual current version of any regulatory guideline, SOP, or study data you want the AI to reference — do not rely on the model’s background knowledge of what a regulation “typically” requires.
- Use explicit instructions to flag inconclusive or ambiguous findings rather than resolving them — ambiguity in your source data should produce ambiguity in the draft, not false certainty.
Related
- Field Role: QA and Compliance Teams
- AI for Deviation and CAPA Documentation
- AI for GDP Audit Preparation
- AI for SOP and Batch Record Review
Last updated August 7, 2026
