The task
A PM hands you a one-paragraph feature idea. Engineering wants user stories with acceptance criteria before they will estimate. You have done this 50 times and you know it always takes longer than it should. AI can take that paragraph and a few persona notes and produce a first-draft story set you then refine, instead of staring at a blank page.
When AI is the right tool
- You write 3+ story sets a week and the format is repetitive.
- The feature is recognizable (CRUD, notifications, search, payments, settings).
- You want a complete first draft to react to, not a from-scratch brainstorm.
- You need stories standardized across squads or contractors.
When not to rely on AI alone
- For novel UX where the story emerges from prototyping. Write the story after, not before.
- For regulated features (medical, financial) where acceptance criteria have legal weight — get a domain SME to review every line.
- For technical-debt or platform work where the “user” is another engineer; the story format does not fit naturally.
What to feed the AI
- The feature idea in 3-5 sentences.
- 1-3 user personas (role, what they care about, what tools they use today).
- Known edge cases and constraints.
- Done definition: what does “shipped” mean for your team.
- Examples of 1-2 well-written past stories so the AI matches your house style.
Copy-ready prompt
Write user stories for this feature.
Feature: {3-5 sentence description}
Personas: {list, with goals}
Edge cases to cover: {list}
Constraints: {auth, perms, data sources, etc.}
Format for each story:
- Title (short, action-oriented).
- "As a [persona], I want [action], so that [outcome]."
- Acceptance criteria as Given/When/Then. Cover happy path + at least 2 edge cases.
- Out-of-scope (1 line) — what this story does NOT cover.
Output:
- 5-9 stories ordered by priority.
- A "we did not cover" list at the end for things you noticed but excluded.Recommended output structure
A clean story set has 5-9 stories ordered by user value, each with a tight title, a one-line “as a / I want / so that,” 3-6 Given/When/Then criteria, and an explicit out-of-scope line. The “we did not cover” tail at the end surfaces gaps for the team to debate.
How to check the output
- Read aloud each acceptance criterion. If you cannot picture clicking through it, it is too vague.
- Are the stories independently shippable? If story 4 cannot exist without story 7, merge or reorder.
- Did the AI invent personas you do not have? Strike those.
Common mistakes
- Stories without acceptance criteria. They get bounced back at planning.
- One giant story that should have been split. Cap at ~3-day estimates per story.
- Skipping edge cases. The AI will only cover them if you list them in the prompt.
Next steps to keep improving
After each sprint, note which stories caused rework or QA escapes. Feed those patterns back: “stories like X tend to miss permission-denied paths — always add a Given/When/Then for unauthorized access.” Your prompt gets sharper every sprint.
Practical depth notes
For Write User Stories With AI: From Feature Idea to Stories With Acceptance Criteria, the difference between a usable AI result and a generic one is the input packet. Give the model the audience, the current draft or raw material, the desired format, the decision you need to make, and two examples of what good and bad output look like. Ask it to preserve facts first, then improve structure or wording second.
After the first response, do a separate review pass. Look for missing constraints, invented details, weak calls to action, and language that sounds plausible but does not match the real situation. The best final output should be easy to use immediately: clear owner, clear next step, and no hidden assumption that someone else has to untangle. A stronger version of this workflow also defines the handoff. Decide who will use the output, what they should do next, and what information would make them reject it. If the deliverable is copy, test whether it has a single clear action. If it is analysis, test whether it separates observation from recommendation. If it is planning, test whether dates, owners, and tradeoffs are explicit enough for someone else to execute. One final check: compare the finished result against the original goal in a single sentence. If that sentence is hard to write, the output is probably polished but unfocused. Tighten the goal, remove decorative language, and rerun only the weak section instead of regenerating the entire piece.
FAQ
- Should the AI write tests too? It can produce test outlines from acceptance criteria — useful for QA scaffolding, not a replacement for real tests.
- How many stories per feature? Usually 5-9. If you get 15, the feature is too big; split it.
- Can it estimate story points? It can guess — do not trust it; estimation is the team’s calibration job.
Related
Tags: #Workflow