Keywordsai for product managersBlogai for workai at workworkplace aiai for professionalshow to use ai at workai productivity
Related searchesai for product managerschatgpt for prdai product discoverychatgpt product managerai for product discoveryhow to use ai for product managers
A PRD is a bet, not a document the model completes
Product managers can use AI to cluster interview notes, turn a messy brief into a PRD shape, and list the questions a sceptical engineer will ask. That is useful labour. It is not the job. The job is the bet: which user, which problem, which trade-off you will accept, and what you will not build. ChatGPT will happily write a vision paragraph, a success metric, and a competitive section. It does not know whether the metric is in the warehouse, whether the user exists in your segment, or whether the trade-off will survive the first steering meeting. Someone named still has to walk in with the bet.
If you cannot name the user, the bet, and the thing you will not build, do not generate the PRD yet. You are still shopping for a story.
Four product jobs ChatGPT can do, and one it cannot
It can cluster feedback you paste into themes with example quotes. It can turn your outline into a PRD skeleton with empty cells for missing evidence. It can draft an interview guide from the hypothesis you already hold. It can list risks and kill-criteria for a pre-mortem. It cannot tell you which bet to take, whether the user will pay, or whether the engineer is right about the platform cost. Give it the notes, the decision the document must support, and the bans: no invented metrics, no unnamed customers, no market figures. Ask it to mark gaps. Then you fill the gaps from research and from the warehouse, not from a paragraph that sounds like a strategy offsite.
| PM artefact | Let ChatGPT | You keep |
|---|---|---|
| Feedback pack | Themes and sample quotes | Which theme is the bet |
| PRD draft | Structure and gap markers | User, metric, and trade-off |
| Interview guide | Questions from your hypothesis | Who you will actually speak to |
| Pre-mortem | Failure modes to discuss | Which risk you will accept |
Draft the PRD from a pack, not from a vibe
Paste only the discovery pack you are allowed to paste: notes, tickets, a metric screenshot, the constraint from engineering. State the reader: engineering, design, or steering. Ban market sizes, persona names, and success metrics you did not supply. Ask for a one-page PRD with [missing] on blank fields. Then open the source. If the draft quotes a user you did not interview, delete the quote. If it sets a target the dashboard cannot measure, delete the target. Keep a small set of saved briefs: feedback clustering, PRD skeleton, interview guide. Edit those, not a blank chat. Product writing is repeating writing. A stable brief is how the team stops arguing with a hallucination about the user.
Role: You are drafting a product artefact I will defend in a review.
Task: Produce the PRD, theme list, or interview guide I specify from the pack I paste.
Context: Reader, decision required, and what they already know.
Constraints:
- Do not add users, metrics, or market figures I did not state.
- If evidence is missing, write [missing] instead of guessing.
- Separate quotes in the pack from inferences.
Output: The artefact, then a list of bets a sceptical engineer would challenge.
Quality checks: Which lines would collapse if the warehouse number is different.If the critique list shows a metric or a persona you did not intend, you have already been saved a bad review. That list is the point of the prompt. Save it in a shared note so the next product manager does not start from a blank PRD. Keep the version that survived contact with engineering, and retire the one that filled every gap with a vision sentence. Five briefs the team uses beat a private trick.
Fake users, fake metrics, and a roadmap that writes itself
Fake users are the quiet failure. A model will name a persona and give them a quote because documents look unfinished without a customer. If you did not hear it, it is not evidence. Fake metrics are worse: a North Star that nobody can query, or a percentage copied from a blog. Roadmaps that write themselves are the third failure. ChatGPT will sequence features because lists want an order. Sequence is a bet about capacity and risk. You still choose what slips. Fluency in the PRD is not discovery. Discovery is contact with a user, a system, and a trade-off you can say out loud without looking at the model.
| Mistake | What it looks like | What to do instead |
|---|---|---|
| What should we build | No pack, no user, no metric | Paste notes and name the bet |
| Invented persona | A named user you never met | Quotes from interviews you ran |
| Dashboard fiction | Increase retention 20 percent | A metric you can open today |
| Roadmap filler | Phase two will delight | One bet and one non-goal |
A PRD that invents a user or a metric is not a draft. It is a faster way to get a room to nod at a bet nobody owns.
Related reading on StudyGrid
Read next: AI as a Thinking Partner AI for Research at Work Writing Someone Will Sign. Those essays sit beside this one. Use them when you need the neighbouring skill, not as a substitute for the check you still have to make.
What to do this week
Take one live bet this week. Cluster only feedback you already have. Draft the PRD with [missing] on blank fields. Check the metric in the warehouse before the review. Keep the brief if the room challenged the trade-off, not the fiction, and share it with the next product manager. Drop it if they still invent a persona. That is how AI for product managers stays beside the work instead of replacing it.