Method
PRD Doctor reads your document against twelve categories a reviewer usually cares about. Each category is a short checklist of things to look for. The score is the weighted sum, and every fix quotes a sentence from your own text. Nothing is generated or guessed on your behalf.
A heading or label such as “Risks” counts only if the section has real content under it.
“Risks: none”, “TBD”, “N/A” and empty sections are detected and capped at 15 for that category.
A single sentence can earn credit in at most four categories: the section it sits in, then its best matches.
A planned interview or an example query does not prove the problem. A PRD about experiments or prioritization does not get credit for those words either: that credit comes from its own plan.
Concrete numbers (limits, counts, dates) and named owners add a small bonus where they matter.
“Fast”, “seamless” or “better” in metrics and acceptance criteria subtract 15. “Everyone” as a target user does too.
A PRD that starts from a solution can't be challenged, prioritised or measured. The problem statement is what the team returns to when scope and trade-offs get argued.
“Everyone” is not a user. Without a specific segment you can't recruit research participants, judge usability, or know whether a metric moved for the right people.
Opinions are cheap. Evidence is what separates a bet from a hunch and lets reviewers decide whether the problem is real before spending engineering time.
Severity decides priority. A frequent, expensive pain justifies a roadmap slot; a rare annoyance doesn't, no matter how elegant the solution.
Unwritten non-goals become scope creep. Explicit boundaries let engineering estimate honestly and let stakeholders disagree early, on paper.
If you can't say how you'll know it worked, you can't tell a win from a miss, and the launch review turns into opinion.
Acceptance criteria are the contract between PM, design, engineering and QA. Without them, “done” is whatever the loudest person remembers.
Every plan has failure modes. Naming them early is cheap; discovering them in production, or in legal review the week before launch, is not.
Most production bugs and support tickets live off the happy path. Edge cases decide whether engineering estimates are real.
Features slip because of what they wait on, not what they build. Unowned dependencies are the most common cause of surprise delays.
When everything is a must-have, nothing is. Ranked requirements are what let the team cut scope calmly when the schedule slips.
A PRD without a test plan is a belief. An experiment turns it into a decision you can defend and reverse.
Points within a category add up to 100, then each category counts for its share of the overall score.
Scoring runs entirely in your browser. PDFs and Word files are read on your device, and your text is never uploaded. Results are kept in your own browser storage so you can edit and re-score. Anonymous usage analytics record counts and scores only, never document text. A shared link or card contains scores, not your PRD.