5 AI prompts I reuse every sprint as a PO
Five copy-ready prompts for research, user stories, acceptance criteria, edge cases and AI agent briefs, each with the check a PO should never skip.
Written by Phan Lam Duy Khoi
Five prompts help a PO draft research themes, user stories, acceptance criteria, edge cases and AI agent briefs faster. Each comes with a check, so the decision stays with the PO.
AI writes the first draft. The PO owns the last word
Every sprint I use AI for the same group of tasks: sorting research, slicing user stories, writing acceptance criteria, finding edge cases and handing work to AI agents. These tasks take time but have a clear structure, so AI drafts them fast.
The problem is that an AI draft always looks fine, even when it is wrong. So every prompt below comes with one check I never skip. The prompt saves time. The check is what makes the output worth using.
- Replace the [brackets] with your real context: product, users, constraints.
- Do not paste personal or internal data unless your company allows it.
- Read the output like a draft from a new colleague: useful, but checked.
Prompt 1: turn messy notes into themes you can trust
Use it when you have a pile of interview notes, customer feedback or support tickets and need to know which problems really repeat.
Here are [N] interview notes and support tickets. Group them into 3–7 themes. For each theme: - a one-line summary - how many sources mention it - 2 direct quotes, with the source ID Put anything that fits no theme in a separate list. Do not invent or reword quotes.
My check: I open every quote and compare it with the raw notes. A theme with one source is a hunch, not a pattern. If a quote is not in the original, I drop that theme and look again.
Prompt 2: slice a big idea into stories that can ship alone
Use it when an idea sounds right but is too big for one sprint. The goal is the smallest piece that still gives users a result they can see.
Feature idea: [idea]. Split it into the smallest user stories that each deliver one outcome a user can notice. For each: "As a… I want… so that…", and why it can ship on its own. Mark which story to ship first and what we learn from it.
My check: a story with no visible outcome for the user (for example, "create the data table") gets merged into another story or cut. AI tends to slice by technical layer; a PO slices by value.
Prompt 3: acceptance criteria someone else can test
Use it once the story is agreed. The goal is criteria clear enough that QA knows what to test without asking again.
Write acceptance criteria for this story in Given / When / Then. Maximum 6. Each one must be testable by someone who did not write it. Flag every vague word (fast, easy, intuitive) and suggest a measurable version. Story: [paste the user story]
My check: I delete any criterion the team will not actually test. Numbers, deadlines and permissions in the criteria must come from a policy or an owner, not from the AI.
Prompt 4: ask AI to break the flow before users do
Use it before you lock the sprint scope. AI finds cases a busy person easily misses.
Act as a QA engineer who wants [flow] to fail. List the ways it could break: empty and error states, permissions, double submits, slow or lost network, time zones, odd input, concurrent edits. Rank by likelihood × impact, with the expected behaviour for each.
My check: I choose which cases go into this sprint. The rest go to the backlog, each with the reason it waits. A long list does not mean we build all of it.
Prompt 5: hand a task to an AI agent without losing control
Use it when an AI agent does a piece of work, such as a UI fix, tests or a prototype. A good brief answers three questions: what done means, what must not be touched, and how we check it.
Task: [what to do]. Done means: [a result I can see]. Do not touch: [files, data, systems]. Check it by: [test or steps]. Stop and ask me before merging, releasing, or anything a customer will see. When finished, report what you changed and what you did not verify.
My check: I read the "did not verify" part of the agent's report first, then use the result myself in the real environment. Green tests are not a working product.
The five checks, in one place
| Task | AI does | The PO checks |
|---|---|---|
| Research | Groups notes into themes, with quotes | Quotes match the source; one source is a hunch |
| Backlog | Slices the idea into small stories | No visible outcome: merge or cut |
| Definition of done | Writes Given / When / Then criteria | Delete what nobody tests; numbers need a source |
| Edge cases | Lists how the flow could break | Choose what goes into this sprint, give a reason for the rest |
| Agent brief | Does the work from the brief | Read what was not verified; try it in the real environment |
AI makes me faster at everything around a decision. The decision itself still belongs to the PO.
Frequently asked questions
Do these prompts work with Claude and ChatGPT?
Yes. They do not depend on a tool. What matters is filling the [brackets] with real context and always running the check before you use the output.
Should I paste customer data into a prompt?
Only if your company's policy allows that tool with that data. If you are not sure, remove names, emails, phone numbers and internal figures first.
Why does every prompt come with a check?
Because AI writes fluent drafts, even when they are wrong. The check keeps the decision with the PO: AI drafts, the PO checks, the PO decides.
Sources to check
Related articles
- Using Claude as a PO: where it helps, what you still do: AI drafts fast. This article shows four mistakes to check and a checklist to use before a document reaches the team.
- Can AI replace a Product Owner? What still needs a PO: A balanced answer for new POs: AI helps with drafting, summarising and suggesting, while accountability for product value and the decisions behind it stays with a person.
- What data should you keep out of AI tools at work?: Six kinds of data a PO should keep out of AI tools, how to filter each piece before pasting, an anonymised complaint example and a ready-to-use checklist.
- Using AI to summarise customer feedback: what to check: AI tools group feedback and name themes fast, but they can invent quotes, lose frequency and drop small groups of customers. This article shows each failure mode and how a PO checks a summary before using it for a decision.