Retrospective and Lessons Learned prompts for Project Manager
Ready-to-use AI prompts for retrospective and lessons learned — written for Project Manager and easy to paste into ChatGPT, Claude or Gemini.
Design a retrospective that surfaces real issues
You are a facilitator planning a retrospective. Team: [team]. Period covered: [period]. What happened: [what happened]. Team dynamics: [team dynamics]. Previous retros produced: [previous retros].
Design a 60-minute session: the format and why it fits this team's dynamics, the opening that makes it safe to be honest, the specific prompts, how to get quiet people talking, and how to convert discussion into a small number of committed actions.
Include how to handle it if the real issue is a person or a leader in the room. End with what to do differently from last time so this retro produces change rather than a list.
Synthesize retro input into actions
Act as a team coach. Here is what the team wrote:
<input>
[input]
</input>
Context: [context].
Cluster the input into themes by root cause rather than symptom. For each theme: what people described, how many raised it, whether it is within the team's control, external, or systemic, and what would actually change it.
Then recommend at most three actions — each with an owner, a specific first step, and how we will know in four weeks whether it worked. Explicitly park the rest, and say which parked items belong to someone outside this team. Flag any theme where the team is describing a symptom of a problem they have not named.
Blameless postmortem of a failure
You are facilitating a postmortem. Incident or failure: [incident]. Timeline: [timeline]. What people knew at each point: [known at each point]. Contributing factors: [contributing factors].
Write a blameless postmortem: the impact in user or business terms, the timeline, the contributing factors traced to systems and processes rather than individuals, why each decision made sense given what was known at the time, what detection or prevention was missing, and the action items with owners.
For every action item, distinguish those that fix this specific failure from those that fix the class of failure — prefer the second. Flag any conclusion that is really hindsight bias, and any place where the timeline has a gap we should fill before publishing.