Write a Feature Spec
Turn a rough feature idea into a clean spec covering behavior, edge cases and dependencies that your team can build from right away.
Last updated: July 2026 · Collective Brain
Good for
- Document a new feature cleanly before the sprint so engineering and design share one picture
- Surface hidden edge cases and error states before they reach the code
- Make dependencies on other systems and teams visible early
The prompt
You are an experienced product manager who writes clear, actionable feature specs for engineering and design.
Context:
- Product: [your product]
- Feature idea: [feature in 1 to 3 sentences]
- User group and goal: [who uses it and why]
Task: Write a complete specification for this feature. Where information is missing, make sensible assumptions and label each one clearly as an assumption.
Structure the output into exactly these sections:
1. Summary: What the feature does and which problem it solves (2 to 3 sentences).
2. User stories: 3 to 5 stories in the format "As a [role] I want [goal] so that [benefit]".
3. Behavior: The happy path step by step, including inputs, system responses and the visible result.
4. Edge cases and error states: Empty states, invalid input, timeouts, concurrent access, permissions. Give the expected behavior for each.
5. Dependencies: APIs, data models, other teams or features that are affected or must be in place first.
6. Acceptance criteria: Testable items as a checklist that define what "done" means.
7. Open questions: What needs to be clarified before building.
Write precisely and plainly. Use bullet points instead of long paragraphs. At the end, ask me up to three follow-up questions if anything important stays unclear.Replace the bracketed placeholders with your own details.
Frequently asked
How detailed does my feature idea need to be in the prompt?
Three or four sentences are enough for a first pass. The more you add about user group, goal and context, the less ChatGPT has to guess. With thin input the model makes assumptions, which it labels for you so you can correct them afterwards.
Can I rely on the edge cases ChatGPT produces?
ChatGPT covers many common cases reliably, but it does not know your technical reality. Treat the list as a strong starting point, not a complete proof. Review it with your engineering team against the actual architecture and add domain-specific cases yourself.
Related
Rather have it done?
Prompts are a start.
Results are our job.
When the prompt should turn into real work that holds up consistently across every channel, we take over. Start free, finish professionally.