Language: English
Start 14-day free trial

Guide

How to write a PRD with ChatGPT from your notes

Updated

ChatGPT or Claude can turn a pile of notes into the first draft of a PRD in minutes, if you ask for a fixed shape. Paste the notes, Slack threads and call notes, and ask for: the problem and for whom, with the evidence under it; one primary user in a specific situation; goals as outcomes and at least two non-goals; one success metric with a baseline and target; requirements ranked must, should and could; and open questions, including every place two notes disagree. Ask it to mark anything it inferred as [assumption] and to leave blanks rather than invent a number. You then fill the blanks and settle the open questions with the team, instead of starting from an empty page.

Short answer

  • Paste the notes; ask for a fixed shape.
  • Problem with evidence, one user, goals and non-goals.
  • One metric, must, should and could.
  • Guesses marked, numbers never invented.

The brief to copy

PRD brief

"Turn these notes into a PRD.

Problem: what a user can't do today, or what it costs them, and the evidence from the notes under it (quotes, ticket counts, lost deals). If there's no evidence, say so.

Users: one primary user in a specific situation, at most two secondary.

Goals as outcomes, and at least two non-goals, including anything the notes discussed and dropped.

One primary metric and one guardrail, each with a baseline and target. If the notes don't give them, write [open: baseline] and [open: target].

Requirements as must, should and could, each one testable. Then open questions (every disagreement between notes, with both positions) and risks.

Mark anything you inferred as [assumption]. Never fill a number, date or customer name that isn't in the notes.

Notes: [paste]"

A worked example

Illustration: a made-up business. A booking app for yoga studios has notes on a waitlist: 32 support tickets in September asking for one, two lost deals that named it, a sales note saying "waitlists should notify by SMS", and a support note saying "studios don't want to pay for SMS". The PRD, in part:

From the PRD

Problem: when a class is full, a studio can't offer the place to someone else when a booking is cancelled, so the spot goes unsold. Evidence: 32 tickets in September; two lost deals named it.
Primary user: a studio manager whose evening classes fill up.
Non-goals: waitlists for private sessions; paid priority places.
Metric: share of cancelled places that are re-booked, [open: baseline], [open: target]. Guardrail: double bookings, [open: baseline], [open: target], with no rise.
Must: a member can join the waitlist of a full class. When a place opens, the first person on the list is offered it for [open: how long]. Should: the studio sees the waitlist for each class. Could: members see their place on the list.
Open question: notify by SMS (sales) or not, because studios don't want to pay for it (support)? Email by default is an [assumption].

The product manager takes the two open questions and the baseline to the next planning meeting, and nothing in the draft claims a number nobody measured.

Before you share it

  1. Check every piece of evidence against the notes.
  2. Fill or keep the [open] marks; don't let a guessed baseline in.
  3. Settle every [assumption] with the person who'd know.
  4. Remove customer names and personal data the PRD doesn't need; see what not to paste.
  5. Save the brief for the next feature.

Next step: review it before the build

A PRD the team builds from deserves the same review every time. Reviewing a product spec with AI before the team builds it covers the checks: the user's path, every screen's states, vague words, limits, the metric and the rollback.

Skills for this

All skills

Questions

Can ChatGPT write a PRD?

It can write the first draft from your notes, if you ask for a fixed shape: the problem with its evidence, one primary user, goals and non-goals, one metric with a baseline and target, requirements as must, should and could, and open questions. Ask it to mark anything it inferred as [assumption] and never to fill a number the notes don't give.

What should a PRD include?

The problem and for whom, with evidence; the users and their situation; goals as outcomes and at least two non-goals; one success metric with a baseline and target, plus a guardrail; requirements ranked must, should and could; and the open questions and risks.

What shouldn't I paste into ChatGPT for a PRD?

Customer names and personal data the PRD doesn't need, and anything your company hasn't approved for AI tools. On a personal plan, turn off model training in settings first; for company work, use your company's account.

AI skill

Have your AI turn messy notes into a PRD

Put this skill in AI Skill Pod and everyone on your team gets it in their Claude, Cursor or ChatGPT (Pro, Business, Enterprise): one invite link, free to start. With PRD from messy notes, each person's AI turns notes into a PRD: the problem with its evidence, the users, goals and non-goals, a metric with a baseline and target, must, should and could requirements, open questions, and anything inferred marked [assumption].

Use this skill in your AI

14 days free · No card needed

Sources

  1. Marty Cagan, SVPG: Assessing product opportunities — first: what problem does this solve, and for whom (read 2026-11-04)
  2. Agile Business Consortium: MoSCoW prioritisation — must, should, could and won't have this time (read 2026-11-04)
  3. OpenAI Help: Data controls — on Free, Plus and Pro personal accounts, model training is on by default and can be turned off; read through search-result text (read 2026-11-05)