Coach the Product Owner With AI, Do Not Write for Them

Coach the Product Owner With AI, Do Not Write for Them

Key Takeaways

  • Accountability Stays Put: The Scrum Guide keeps the Product Backlog with the Product Owner, whatever tools are involved.
  • The Secretary Anti-Pattern: A Scrum Master who writes every item with AI becomes a scribe, not a coach.
  • PO Drafts, AI Questions, SM Coaches: Use AI to challenge the Product Owner's thinking, never to replace it.
  • Five Coaching Prompts: Prompts that expose gaps instead of quietly filling them in for you.

It takes about ten seconds to have an AI write a user story. That speed is exactly why Scrum Masters get pulled into doing it.

It starts as a favour during a busy sprint. Then it becomes the way refinement works. Then one day you notice that you are the only person who fully understands the backlog, the Product Owner has stopped thinking about it, and every decision waits on you.

This guide is part of our hub on AI for Scrum Masters. For the full comparison of tools, read the master guide: ChatGPT vs Claude vs Gemini vs Copilot for Scrum Masters.

What the Scrum Guide Says

The Product Owner is accountable for effective Product Backlog management. The Product Owner may delegate the work to others, but remains accountable for it.

The Scrum Master's accountability includes helping the team find techniques for Product Goal definition and Product Backlog management.

Read those two statements together and the division of labour is clear. The Product Owner owns the backlog. The Scrum Master helps the Product Owner get better at owning it.

Using AI to write the items on the Product Owner's behalf quietly inverts that arrangement. Nothing in the framework stops you, and nobody will complain in the short term, which is what makes it such an easy habit to acquire. We describe the wider pattern in Will AI Replace the Scrum Master?

The Rule: The PO Drafts, the AI Questions, the SM Coaches

WhoDoes thisWhy
Product OwnerWrites and decides every backlog itemThe Scrum Guide keeps accountability for the Product Backlog with the Product Owner.
AIAsks questions, plays the sceptic, suggests slicing optionsIt sparks thinking without owning the outcome.
Scrum MasterCoaches the Product Owner on techniqueThe Guide lists helping find techniques for Product Backlog management as part of the role.

The test is simple. After the session, is the Product Owner more capable, or just more dependent on you?

The Difference in One Example

Compare two prompts run on the same rough item.

Writing for them: "Turn this into a properly formatted user story with acceptance criteria."
Coaching them: "Ask me five questions that expose what is missing from this item. Do not rewrite it."

The first produces a polished item and no new understanding. The second produces an uncomfortable list of gaps and a Product Owner who now knows what they did not know. Only one of those compounds over time.

Five Prompts That Coach Instead of Write

Have the Product Owner paste their own draft item and run these. The phrasing matters: each one explicitly forbids the model from doing the work.

The Coaching Prompt Library

Have the Product Owner paste their own draft item into one of these. Every prompt explicitly forbids the model from doing the work, because left to itself a model would rather help than interrogate. Copy them, keep them somewhere the Product Owner can reach without you, and that is the whole point.

1. Expose the gaps

Ask me five questions that expose what is missing from this backlog item. Do not rewrite it. Do not suggest wording. After the questions, say which single gap would cause the most rework if we ignored it. ITEM: [paste your draft item]

2. Play the sceptical Developer

Act as a sceptical Developer who has to estimate this item tomorrow. List every question you would need answered before you could give an estimate you would stand behind. Rank them by how much they would change your estimate. Do not rewrite the item and do not estimate it yourself. ITEM: [paste your draft item]

3. Find the slices

List three ways to slice this backlog item so that each slice delivers value on its own. For each slicing option give: what the first slice would deliver, what it deliberately leaves out, and the trade-off we accept by starting there. Do not write the items. We will do that. ITEM: [paste your draft item]

4. Hunt the ambiguity

Highlight every ambiguous or unmeasurable word in these acceptance criteria. For each one, explain why it is ambiguous and what question I should ask to resolve it. Pay particular attention to words like quickly, seamlessly, intuitive, appropriate, relevant and easily. Do not rewrite the criteria. ACCEPTANCE CRITERIA: [paste your criteria]

5. Name the riskiest assumption

What is the riskiest assumption inside this backlog item? Explain why it is risky, what happens if it is wrong, and how we could test it in a single day without building the feature. Give one alternative assumption we may be overlooking. ITEM: [paste your draft item]

If the model rewrites the item anyway, say so in the same chat: “You rewrote it. Ask questions only.” Restating the boundary works better than hunting for a tool that never needs to be told.

Prompt four is the one that changes behaviour fastest. Words such as "quickly", "seamlessly", "intuitive" and "appropriate" survive in acceptance criteria because they sound like requirements while committing to nothing. Seeing them highlighted a few times teaches a Product Owner to catch them without help.

To pressure-test the idea itself rather than the wording, you can also run it past simulated users. We explain how in Using Synthetic Users to Validate Products.

The 15-Minute Session Checklist

Run this alongside the Product Owner, on their screen, with their hands on the keyboard. The hardest item is the fourth one.

Before the session

During the session

After the session

The test that matters

Signs the Coaching Is Working

  • The Product Owner starts answering the model's questions before it asks them.
  • Items arrive at refinement with the obvious gaps already closed.
  • Developers ask fewer clarifying questions and different ones.
  • You are invited to fewer refinement conversations, not more.

When the Product Owner Wants You to Write Everything

Sometimes the Product Owner is drowning rather than lazy, and the distinction matters for how you respond.

Say it plainly: "I will sit with you while you draft, and we will use AI to challenge it. I will not write the items, because you are accountable for them." Said once, kindly and without apology, this is almost never taken badly.

Then look at why they are overloaded. Too many stakeholders, no real decision rights, three products on one pair of shoulders and no time to think are organisational impediments. Raise them. Do not absorb them. Absorbing an impediment makes it invisible, which guarantees it will never be fixed.

What to Do When a Manager Asks You to Write the Backlog

This request usually comes with good intentions and a delivery deadline attached. Answer at the level of the problem rather than the task: explain that you can make the Product Owner faster at producing good items, or you can produce items yourself and become a single point of failure, and that the first option costs more this month and much less every month after. If the answer is still no, do it, note it as an impediment, and set a date to revisit it.

Which AI Is Best for Backlog Refinement?

We are giving the four tools the same draft item and scoring them on question quality, faithfulness to the draft, the tendency to rewrite instead of question, and invented detail. Results will appear on the master guide with the plan name and test date once our first test round is complete.

One behaviour is worth knowing about in advance. Most models would rather help than interrogate, so they drift toward rewriting even when told not to. If that happens, restate the constraint in the same chat: "You rewrote it. Ask questions only." Repeating the boundary works far better than searching for a tool that never needs to be told.

Do not paste confidential roadmap details into a personal account. Read what you can safely paste first, and check whether an approved work tool is available in which AI fits your stack.

Tools and prices change every quarter. The judgement needed to use them well does not.

Build That Skill With AgileWoW AI for Scrum Masters

Three Anti-Patterns to Watch For

  • The overnight backlog. Forty AI-generated items appear between sprints. They are consistent, well formatted and disconnected from any conversation with a user, and refinement becomes an editing exercise.
  • The acceptance-criteria factory. Every item arrives with eight criteria because the model produced eight. Volume replaces thought, and Developers stop reading them.
  • Refinement by chat log. The Product Owner pastes the model's output into the item verbatim. Nobody, including the Product Owner, can explain why the item says what it says.

All three share a symptom: the backlog gets tidier while the team's shared understanding gets thinner. Tidiness is easy to see in a tool and understanding is not, which is exactly why this goes unnoticed for so long.

Conclusion

AI makes it easy to do your Product Owner's job for them. That is precisely why you should not.

Use it to make them sharper, ask the question they were avoiding, and let them own the result. The backlog you did not write is worth more than the one you did.

Next Step: coaching the Product Owner is one relationship. The Sprint Review is where your stakeholders meet the result. Read AI for Sprint Reviews and Stakeholder Updates That Land.

About the Author: Ayush Bisht

Ayush Bisht is a Content Engineer and AI Tools Specialist at AgileWoW, focused on creating smart and scalable digital experiences through AI-powered content solutions.

Frequently Asked Questions (FAQ)

Should a Scrum Master write user stories with AI?

Occasionally helping is fine, but making it a habit is an anti-pattern. The Scrum Guide keeps accountability for the Product Backlog with the Product Owner, and the Scrum Master's role is to coach on technique rather than to supply the output.

Who is accountable for the Product Backlog?

The Product Owner. According to the Scrum Guide, the Product Owner may delegate the work to others but remains accountable for effective Product Backlog management.

Is it an anti-pattern for the Scrum Master to write backlog items?

It becomes one when it is routine. It weakens the Product Owner's ownership and turns the Scrum Master into a scribe. Coaching the Product Owner to draft and refine items is the better use of the role.

Can AI split large backlog items?

Yes, it can suggest slicing options with trade-offs. The Product Owner and Developers should choose between them, because they hold the customer and technical context that the model lacks.

How do I coach a Product Owner who wants me to write everything?

Offer to work alongside them while they draft, use AI to ask challenging questions, and explain plainly that they are accountable for the backlog. If they are genuinely overloaded, raise that as an impediment instead of absorbing the work.

What prompts help a Product Owner improve a backlog item?

The ones that forbid rewriting. Ask the model to list what is missing, to act as a sceptical Developer preparing to estimate, to flag ambiguous words in the acceptance criteria, or to name the riskiest assumption and how to test it within a day.

Can AI replace backlog refinement sessions?

No. Refinement is where Developers, the Product Owner and sometimes stakeholders build shared understanding of what is being asked for. AI can improve what each person brings to the session, but a well-written item is not a substitute for the conversation that produced it.

Should the Product Owner use AI directly, without the Scrum Master?

Ideally yes, and that is the goal of the coaching routine in this guide. A Product Owner who runs these prompts independently is exactly the outcome you want, and it frees you for the organisational impediments that only you can raise.

Does AI-generated acceptance criteria improve quality?

Sometimes, and often it just adds volume. Criteria written without a conversation tend to be plausible rather than correct, and long lists discourage the people who need to read them. Use AI to test criteria for ambiguity rather than to produce them.

What if my Product Owner refuses to use AI at all?

That is entirely their call, and the coaching techniques still work without it. The prompts in this guide are really just good refinement questions, and you can ask them yourself. The tool is convenient, not essential.

References and Sources