Give five stakeholders one clear PRD, informed by several writing perspectives

Fix the scope before improving the sentence

Write the current decision in plain terms: who needs what outcome, which conditions apply, what is out of scope and which questions remain unanswered. Mark a proposed decision differently from an approved one. Otherwise an elegant rewrite may quietly turn a suggestion into a requirement.

Atlassian describes a PRD as a shared account of product purpose, features, user needs and success criteria. Use that shared purpose as the constraint on AI variation. Stakeholder-specific explanations should illuminate the same requirement, not create five versions of the product that cannot all be delivered.

A fictional requirement that invites the right questions

Consider a fictional reporting product with this draft: Users should be able to export reports easily. The approved scope is narrower: workspace owners can export the currently filtered table as a UTF-8 CSV file. Scheduled exports are excluded. File-size limits, empty results and failure behavior are still undecided. These are illustrative requirements, not Colay features.

A clearer core statement is: A workspace owner can export the rows matching the active table filters as a UTF-8 CSV file. Put the exclusions and unanswered questions next to it. Do not let a model replace the missing performance decision with an invented duration.

Request eight lenses, then consolidate them

Use selected available agents in Ask separately to inspect their initial interpretations. Eight requested lenses do not require eight independent models. Keep any disagreement that points to a real product choice. For example, whether an empty table produces a header-only file is a decision for the team, not a stylistic preference to average away.

Review this PRD excerpt: [text]. Approved decisions: [facts]. Proposals: [not yet approved]. Out of scope: [list]. Open questions: [list]. Produce eight formulations emphasizing user outcome, concise scope, testability, permissions, failure behavior, performance, support needs and executive context. Every version must preserve the same approved scope. Do not invent technical limits or resolve open questions silently. For each version, identify one ambiguity it exposes. Then propose one canonical requirement, example acceptance cases, and a decision list with suggested owners. Mark every new suggestion as a proposal.

Build the review packet around one requirement

For the fictional export requirement, an approved case can check that an owner receives a CSV containing rows matching active filters. A non-owner's expected interface and an empty-result response remain questions unless the team has decided them. This distinction prevents an AI-generated test list from becoming accidental scope.

1. Choose one core formulation and compare it with the approved decision list. Check roles, conditions, outputs and exclusions word by word.
2. Add example acceptance cases for the decisions already made. Label cases involving unresolved behavior as questions rather than agreed tests.
3. Separate wording edits from scope proposals. A new row limit, permission rule or recovery path needs a product decision even when the sentence reads better.
4. Send reviewers the canonical text, meaningful changes and open questions with owners through your existing review process.

Use the macOS shortcut for a focused wording pass

Keep the PRD open and invoke Colay with ⌘⇧K by default, or your configured shortcut in Settings. Paste only the excerpt and permitted context needed for the review. The overlay opens a drafting workspace; it does not read the document automatically or submit the PRD for approval.

After the review, record which comments changed wording and which changed the product decision. Update the canonical requirement and its examples together. Reuse the prompt on a difficult paragraph when needed, rather than repeatedly rewriting the whole PRD. A useful result is a smaller, clearer set of real decisions for the team to make.

Questions, answered