Library / ChatGPT, Claude & prompting

Prompt engineering for IT: a simple framework and 10 prompts

A repeatable way to write prompts that get useful answers, with ten ready-made prompts for common IT tasks. Works with Copilot, ChatGPT and Claude.

Guide Beginner 1 Nov 2026 By Tariq Hanif

The RCTF framework

Good prompts usually include four things:

  • Role: who the AI should act as, and who it’s helping.
  • Context: the environment, versions and what you’ve tried.
  • Task: exactly what you want.
  • Format: how you want the answer (steps, table, script, email).

Weak: Fix my Intune issue.

Strong: Act as a senior endpoint engineer helping a junior admin. Windows 11 devices enrolled via Autopilot show "Not compliant" for BitLocker even though BitLocker is on. We use Intune compliance policies. Give the five most likely causes in order, how to check each, and the fix. Use a numbered list.

Ten prompts to copy

  1. Explain a concept: Explain [topic] to someone in their first year in IT, with one real-world example and three common mistakes.
  2. Error message: Here is an error message and what I was doing. List likely causes in order and how to confirm each.
  3. Write a script: Write a PowerShell script that [task]. Include comments, error handling and a -WhatIf option. Use Microsoft Graph PowerShell where possible.
  4. Review a script: Review this script for bugs, security issues and destructive actions. Suggest a safer version.
  5. KQL query: Write a KQL query for Microsoft Sentinel that finds [behaviour] in the last 7 days, and explain each line.
  6. Runbook: Turn these notes into a runbook with prerequisites, steps, rollback and validation.
  7. User email: Write a short, friendly email to staff about [change], avoiding technical jargon.
  8. Change request: Draft a change request for [change] with risk, impact, rollback and testing.
  9. Study plan: Create a 6-week study plan for [certification] for someone working full-time, with weekly labs.
  10. Interview practice: Ask me 10 interview questions for a [role] position, one at a time, and give feedback on each answer.

Always remember

AI can be confidently wrong. Check commands against official documentation and test in a lab before production.