Library › Agents & Automation AI › Product managers

Write custom GPT instructions for a release-notes helper

  • For Product managers
  • You practice Setting limits and rules
  • Time about 5 minutes

About this exercise

You're setting up a team custom GPT that turns this week's merged tickets into customer release notes; the test below runs your instructions on real tickets. It must draft one note per shipped customer-facing change, skip internal and reverted work, never name a customer or ticket ID, and never publish: it asks you to approve instead. Output is a plain-text numbered action log: draft_note lines, then one ASK_HUMAN line, then a DONE line.

Common mistakes when you ask AI for this

  • Not showing the log format. Left alone, the AI writes a friendly report with headings instead of a checkable action log. Try: Show it literally: "Format every line like 1. ACTION: tool(args), then ASK_HUMAN: and DONE: lines. Plain text."
  • Not saying what counts as shipped. The AI includes everything merged, including the reverted dark mode and internal plumbing. Try: Add: "Only shipped, customer-facing changes. Skip internal work and anything reverted."
  • Leaking internal details. The AI copies ticket IDs and the requesting customer's name straight into public notes. Try: Add: "Never include ticket IDs or customer names."
  • Letting it publish. Given a publish action, the AI assumes finishing the job means using it. Try: Add: "Never call publish_changelog. Ask me to approve with ASK_HUMAN, then DONE."

What a strong prompt includes

  • Skips internal and reverted work
  • Keeps IDs and names out
  • Forbids publishing
  • Makes it ask before anything risky
  • Shows the action-log format

Practice this now ↑

Related exercises