Loading page
About this tool
Build a PM assistant system prompt for specs, user stories and tradeoffs, with a real prioritisation framework wired in.
This builder assembles a system prompt for a product-management assistant that writes PRDs with explicit non-goals, user stories in the Connextra format meeting the INVEST checklist with Given/When/Then acceptance criteria, and backlog priorities scored with a real framework — RICE, ICE, MoSCoW or WSJF — with the arithmetic shown. It is for product managers and founders who want AI drafting help that argues from evidence instead of inventing metrics.
Open Product Manager System Prompt Builder on AltFTool — it loads instantly in your browser.
Add your input to the workspace.
Adjust the options until the result looks right.
Copy or download the output and put it to work.
RICE (Intercom), ICE, MoSCoW and WSJF (SAFe) are written into the prompt with their actual scoring formulas.
One-click rules force assumptions to be labelled and ban invented user research, quotes and metrics.
Every spec must name non-goals and define success as a measurable number with a baseline and date.
RICE, created at Intercom, scores each item as (Reach × Impact × Confidence) ÷ Effort. Reach is users affected per period, Impact is rated 0.25 / 0.5 / 1 / 2 / 3, Confidence is 50%, 80% or 100%, and Effort is in person-months. Higher scores rank higher; the value of the exercise is mostly in arguing about the four inputs, which is why this builder makes the assistant show its arithmetic.
INVEST is Bill Wake's checklist for a good user story: Independent, Negotiable, Valuable, Estimable, Small and Testable. A story failing the checklist — say, too big to estimate or with no testable outcome — should be split or rewritten. The builder pairs INVEST with the Connextra template ('As a …, I want …, so that …') and Gherkin Given/When/Then acceptance criteria.
A working minimum: the problem and evidence for it, goals with measurable success metrics, explicit non-goals (what is out of scope), user flows or scenarios, and open questions. Non-goals are the most commonly skipped section and the biggest source of scope creep, so this builder can make a spec without them count as unfinished.
Give it two hard rules: never invent research, metrics, quotes or competitor facts — ask or leave an open question instead — and label every unbacked claim as 'Assumption:'. These are the two most important rules in this builder and it warns you if the first is disabled. Treat AI-drafted numbers as placeholders until a human sources them.