ChatGPT prompt for a postmortem report

Create a blameless postmortem report from the information below. Do not infer missing events, people, system behavior, customer impact, or root causes. Ask up to 3 clarifying questions only if a required input is missing.

Incident details: [INCIDENT NOTES]
Timeline evidence: [TIMELINE EVIDENCE]
Impact and detection: [IMPACT AND DETECTION]
Contributing conditions: [CONTRIBUTING CONDITIONS]
Action-item owners and constraints: [OWNERS AND CONSTRAINTS]
Audience and report date: [AUDIENCE AND REPORT DATE]

Produce a report in this exact structure:
1. Title, incident date, severity, authors, and status.
2. Executive summary in 100–150 words: what happened, who or what was affected, duration, and current state.
3. Customer and business impact, separating confirmed facts from estimates.
4. Detection and response: how the issue was found, what alerts or signals worked, and what delayed recognition.
5. A timestamped timeline in chronological order. Mark uncertain times as “approximate” and identify gaps.
6. Technical or process root causes using a causal chain, distinguishing root causes from contributing factors and triggers. Do not blame individuals.
7. What went well, what made response harder, and where safeguards failed or were absent.
8. Corrective actions in a table with action, intended risk reduction, owner, due date, priority, and verification method. Actions must be specific enough to close, such as adding a named alert with a threshold or testing a rollback path.
9. Open questions and follow-up review date.

Before finalizing, check that every causal claim is supported by the supplied evidence and flag unsupported claims. Also check that each action has one accountable owner, a realistic due date or “TBD,” and a measurable verification method.

Draft an executive incident update using [CURRENT FACTS], [CUSTOMER IMPACT], [RESPONSE STATUS], and [NEXT UPDATE TIME]. Write 150–220 words with: current status, confirmed impact, actions underway, known unknowns, and the next update commitment. Do not speculate about cause or restoration time. Separate confirmed facts from estimates. Ask up to 3 questions only if a required input is missing. Before answering, check that every number and time appears in my inputs and that no unconfirmed cause is presented as fact.

Review these proposed incident actions: [ACTION ITEMS]. Use this context: [INCIDENT SUMMARY] and [TEAM CONSTRAINTS]. Return a table with each action, the failure mode it addresses, whether it prevents, detects, mitigates, or improves recovery, a measurable acceptance test, owner gaps, and priority. Identify vague actions such as “improve monitoring” and rewrite them into testable work. Ask up to 3 questions only for missing inputs. Check that each recommendation links to an evidenced failure mode and does not assign invented owners.

Write a customer incident notice from [INCIDENT FACTS], [AFFECTED CUSTOMERS], [RESOLUTION], and [SUPPORT PATH]. Produce a subject line plus 120–180 words in plain language. State what customers experienced, the affected window, the current status, and any action they need to take. Do not include internal blame, unverified causes, or promises not supported by my notes. Ask up to 3 questions only if needed. Check that the dates, times, and scope match the supplied facts.