AI Prompt Guide

Artificial Intelligence tools respond to prompts—the specific instructions you provide. To ensure your interactions are clear, accurate, and secure, MSU IT recommends the GCSE Framework as your foundational standard, supplemented by advanced prompting frameworks for complex tasks. 

The GCSE Framework 

  1. Goal (The Action)

Begin your prompt with a specific request using a clear action verb (summarize, draft, analyze, brainstorm). 

  • Less effective: “Tell me about this policy.” 
  • More effective: “Summarize this policy in 5 bullet points for department administrators. Focus specifically on compliance deadlines.” 
  1. Context & Persona (The Audience)

Tell the AI who it should “be” and who the message is for to ensure the right tone. 

  • Role: “Act as an MSU academic advisor…” 
  • Audience: “…writing for incoming first-year students.” 
  • Tone: “Use a welcoming yet professional tone.” 
  1. Source (The Information)

Direct the AI to use specific data or documents to prevent “hallucinations” (made-up facts). 

  • Reference Data: “Use the attached PDF of the faculty handbook…” 
  • Example: “Draft a FAQ based on the meeting transcript provided below.” 
  1. Expectations (The Format)

Define exactly how you want the result to be structured. Don’t settle for a wall of text. 

  • Comparison of Tables: “Provide a table with columns for Feature, Cost, and Security.” 
  • Structural Constraints: “Write a 100-word executive summary with bullet-point action items.” 
  • Sequenced Workflows: You can request a “Chain of Outputs” to handle complex tasks in one go. Example: “Please provide the following in order: 1. A summary of the policy changes, 2. A draft announcement email for staff, and 3. A 3-step checklist for implementation.” 

Advanced Prompting Frameworks 

For complex research or multi-step logic, leverage these advanced prompting frameworks. 

  • Logic & Reasoning: Use Chain-of-Thought (CoT) by asking the AI to “think step-by-step” to improve accuracy in math or logic. For exploring multiple solutions, try Tree of Thoughts (ToT). 
  • Structure & Persona: Use COSTAR to ensure professional-grade context and specific formatting. 
  • Action-Oriented: Use ReAct for workflows where the AI must reason and then perform a specific task or search. 

Iterate & Refine 

Prompting is a conversation. If the first result isn’t perfect, use these two methods to polish it: 

Refine Through Dialogue 

Ask the AI to adjust its own work: 

  • “This is too long; condense it into two paragraphs.” 
  • “Change the tone to be more encouraging for students.” 
  • “Add a section explaining the ‘Why’ behind this change.” 
  • “Include your sources along with your results.” 

Use an AI Prompt Refiner 

If you’re stuck, ask the AI to be your consultant. Use this “Meta-Prompt“: 

“I want to [Insert Goal]. My current prompt is: ‘[Insert Your Prompt]’. Please critique this using the [your preferred framework] and suggest a more effective version.” 

Prompting for a Prompt (The “Golden Prompt” Technique)  

If you’ve refined a complex task through a longer back-and-forth and landed on a result you love, you can make it repeatable by asking the AI to turn your conversation into a single “reuse-ready” prompt for next time. For example: 

“We’ve reached a strong final result. Based on our conversation and the final output, write one comprehensive prompt I can paste into a new chat to reproduce this same result in a single step.” 

Responsible Prompting at MSU 

  • Protect Data: Use MSU-approved tools with enterprise protection. Do not enter any sensitive or university data, especially FERPA or HIPAA protected data, into unapproved AI tools.
  • Verify Accuracy: AI can be confidently wrong. Always cross-reference facts, dates, and names with sources. Never assume the AI generates 100% accurate results. If a result is incorrect, investigate the cause and try to fix it:
    • Fix or exclude data: Correct flawed source material or tell the AI to ignore specific, inaccurate sections.
    • Anchor hallucinations: If the AI “makes things up,” provide more rigid, factual data to keep it on track. 
  • Human Review: You are responsible for the final output. Always review AI-generated content for accuracy before sharing.