AI prompts for sprint retrospectives: a practical guide for Scrum Masters
Seven copy-ready AI prompts for retrospectives: prepare, group cards, find themes, draft actions and write the summary. Works with ChatGPT, Claude or Copilot.
AI will not run your retrospective for you, and it should not. The conversation between people is the point. But a chat assistant such as ChatGPT, Claude or Copilot is very good at the slow parts around the meeting: picking a format, grouping forty sticky notes, spotting a theme that keeps coming back, and turning a messy discussion into three clear actions and a short summary.
This guide gives you seven prompts that work, the order to use them in, and the line you should not cross.
The workflow in one paragraph
Run the retro as usual on a shared board. When the team has written and voted, export the board as text, paste it into your AI assistant with one of the prompts below, and read the answer back to the team as a suggestion to discuss, not a verdict. In RetroLift, open the board menu and choose Copy to clipboard: you get every column, card, vote count, comment and action item as Markdown, without the names of who wrote what.
Before the retro
1. Pick a format that fits the sprint
I facilitate a retrospective for a team of [number] people.
This sprint: [two or three sentences, e.g. "we missed the release date because of a late API change; morale is OK but people are tired"].
Suggest three retrospective formats that suit this situation.
For each, give the columns, why it fits, and one opening question.
Give the assistant the mood of the sprint, not only the facts. "Tired but proud" leads to a different format from "frustrated with another team".
2. Write a check-in question
Write five short check-in questions for the start of a retrospective.
The team is [remote / in the office], the sprint was [hard / calm / mixed].
Each must be answerable in one word or a number, and safe for work.
During the retro
3. Group the cards
Below is the export of our retrospective board.
Group the cards into 3 to 6 themes. For each theme give a short name, the cards that belong to it (quote them exactly), and the total votes.
Do not invent cards and do not drop any. List cards that fit no theme separately.
[paste the board here]
Asking for exact quotes and "do not drop any" keeps the assistant honest. Check the groups with the team: a wrong grouping is a good conversation starter in itself.
4. Find what is underneath
From the same board, what are the two or three underlying causes that could explain several cards at once? For each, name the cards that point to it and one question the team could discuss to check whether it is true.
This is the prompt that saves the most time. Present the answer as questions, because the team knows things the board does not say.
After the discussion
5. Turn discussion into action items
Here are our notes from the discussion: [paste notes or the top themes].
Draft at most three action items. Each must be specific, have a suggested owner role (not a name), a due date within the next sprint, and a way to tell next retro whether it worked.
Three is a deliberate limit. Teams that leave with ten actions usually finish none.
6. Write the summary for the team chat
Write a short summary of this retrospective for our team chat:
what went well, the main theme we discussed, and the action items with owners. Under 120 words, plain language, no emojis.
[paste the board and the agreed actions]
7. Look across several sprints
Below are the exports of our last [3–5] retrospectives, oldest first.
Which problems keep coming back? Which actions did not seem to change anything?
Quote the cards that support each point.
Keep the exports in a folder after each retro and this becomes the most useful prompt of all: it shows the team its own patterns.
What not to give the AI
- Check your company's AI policy first. Many organisations allow only approved tools for internal information.
- Leave out names and personal details. The RetroLift export already omits who wrote each card; also remove names inside card text.
- Do not let it judge people. Prompts like "who is the problem on this team" break the safety that makes a retro work.
- Do not skip the conversation. If the summary is written before people have talked, the retro has become a report.
A 45-minute retro with AI
| Time | Step | AI's part |
|---|---|---|
| Before | Choose format and check-in | Prompts 1 and 2 |
| 0–5 min | Check-in | None |
| 5–15 min | Silent writing and voting | None |
| 15–20 min | Group cards | Prompt 3, checked by the team |
| 20–35 min | Discuss top themes | Prompt 4 as questions |
| 35–45 min | Agree actions | Prompt 5 as a first draft |
| After | Share summary | Prompt 6 |
Frequently asked questions
Can AI facilitate a retrospective?
It can help a facilitator prepare, group cards and write up actions, but it should not replace the facilitator. Trust, timing and reading the room still need a person.
Which AI tool is best for retrospectives?
Any capable chat assistant works with these prompts, including ChatGPT, Claude and Microsoft Copilot. Use the one your organisation has approved.
Is it safe to paste retro notes into ChatGPT?
Only if your company allows it. Remove names and anything confidential first, and prefer a business account that does not use your data for training.
How do I get my board into the AI tool?
Export it as text. RetroLift, a free retro board by AIYUN, copies the whole board as Markdown in one click, including votes, comments and action items.