AI PRD critique
Eleven passes. One sharper PRD.
Critique reads your draft like a senior Product Manager. It runs eleven distinct analytical passes across the dimensions that matter, then lands findings as comments on your spec. The goal is not a longer document. It is a sharper one.
Before & after
From a draft you hope is ready to one you know is.
Before Thinkr
- You reread your own draft and miss the gaps you are too close to see.
- Reviewers spend the meeting finding problems you could have caught alone.
- Feedback arrives as vague “considerations,” not fixes you can act on.
- Severity is a matter of opinion, so the real blockers get buried.
With Thinkr
- Eleven passes read the draft the way a senior reviewer would.
- Blockers, majors, and minors surface before anyone else opens the doc.
- Every finding lands with a suggested rewrite you can accept inline.
- Severity is assigned the same way every time, so you fix what matters first.
The eleven passes
Each pass focuses on a single failure mode that wastes the most reviewer time when it slips through. Findings come back with severity, suggested rewrites, and a pointer to the section that needs work.
- Context Summary: what kind of work is this, and how should that shape the review?
- PRD Completion: each section checked against a template — complete, partial, or missing.
- Initiatives Evaluation: worth doing, right solution, build now, secure & responsible.
- Clarity of Thinking: problem, goals, stories, UX flow, and metrics — with sharper rewrites.
- Engineering Readiness: functional and non-functional requirements, integrations, acceptance criteria.
- User Flow Coverage: every story mapped to flows, with the scenarios the PRD missed.
- Edge Case & QA Drill: edge cases across user, system, data, and security dimensions.
- AI-Specific Readiness: model needs, cost, behavior, and evaluation (when the PRD is AI-related).
- Billing & Commercialization: pricing, billing-system impact, and finance dependencies (when relevant).
- Domain Gap Analysis: the missing parts and wrong assumptions specific to your domain.
- Synthesis & Final Verdict: the top risks, the top improvements, and a deterministic 0–100 score — broken down by dimension with a readiness band, so you can see what dragged the number down.
Findings, not commentary
Critique does not return a list of "considerations." It returns findings: classified, prioritized, with suggested rewrites. Drop them straight into the PRD as resolved comments, or push back on them through the comment thread.
- Each finding has severity (blocker / major / minor) and a category
- Deterministic structural checks (generic personas, missing MoSCoW priority, untestable acceptance criteria) run without the model, so they flag the same way every time
- Suggested rewrites you can accept inline
- Findings land in the PRD as resolved comments so the review history is preserved
- Push back on a finding to refine the critique without rerunning all 11 passes
When to run it
Run it once when you finish the first draft. Run it again before peer review. Run it before publishing to stakeholders. It costs you a minute. It saves you a meeting.
- Manual trigger from the Draft step
- Auto-triggers before stakeholder publish (configurable)
- Re-run after edits to confirm the blockers are actually resolved
How it compares
The AI PRD review that actually publishes its rubric
Most tools tell you they review your spec. None of them show you the checks. Here is what each one actually publishes.
| Publishes its full rubric? | How it reviews | |
|---|---|---|
| Thinkr | Yes — 11 named passes | Severity-classified findings with suggested rewrites, on the line they concern |
| ChatPRD | No — black box | “CPO-level” coaching; the criteria are not stated |
| PMPrompt | No | Names three areas (completeness, clarity, effectiveness); no enumerated checks |
| Centercode Draft Doctor | No | Flags some failure modes inline; no full rubric, no named author |
Competitor details from publicly listed pages, June 2026. Reflects what each tool publishes about its review process.
Questions
AI PRD review, answered
- What is a PRD review?
- A structured read of a product spec before it ships — looking for what is missing or weak (unframed problems, unmeasurable metrics, uncovered edge cases, readiness gaps) rather than polishing what is already written. The output is findings you can act on, sorted by severity.
- Can AI review a PRD?
- Yes, and it is a different job than writing one. An AI reviewer reads your draft against a fixed rubric and returns flagged gaps with suggested fixes. It catches the mechanical failures consistently — missing acceptance criteria, vague metrics, absent error states. The judgment about whether you are solving the right problem still needs you.
- How is a PRD reviewer different from a PRD generator?
- A generator writes a draft from a prompt. A reviewer reads the draft you already have and tells you where it breaks. Most AI PRD tools generate; the review is the step almost none of them run.
- What does the 11-pass critique check?
- Context Summary, PRD Completion, Initiatives Evaluation, Clarity of Thinking, Engineering Readiness, User Flow Coverage, Edge Case and QA, AI-Specific Readiness, Billing and Commercialization, Domain Gap Analysis, and a Synthesis that lands the final 0–100 score.
- How is the critique score calculated?
- The headline is a deterministic 0–100 — your PRD's score of record — with a per-dimension breakdown beneath it: Strategic Clarity, Engineering Readiness, User Flow Coverage, Domain & Risk, and Completeness, each scored 0–100 with a rationale and a readiness band (Excellent to Not Ready). It is computed from the pass results and findings rather than a model's gut feel, so it is reproducible and you can see exactly which dimension pulled the score down. Hover any dimension to see its weight.
- Is my PRD kept private?
- Your drafts are never used to train AI models, and you can delete a workspace and its sources at any time.
Methodology and trust
Who built the rubric, and is your PRD private?
The 11-pass critique comes from a simple frustration: every PRD I shipped went out a little foggy, and my team caught the gaps in review a sprint too late. So the passes are the checks a senior reviewer would run, named in the open, so you can audit what gets checked instead of trusting a black box.
Galang Aulia, FounderMore about the team →
Your data stays yours. Your drafts are never used to train AI models. The stack runs on Google Gemini, where customer content is excluded from model training under its enterprise terms, and you can delete a workspace and its sources anytime.