Creating strong prompts for AI "mentors"
- Katie Capella
Creating an AI “mentor” in mentorAI requires users to carefully craft a set of prompts. While using an LLM like ChatGPT, Claude, or Gemini to create these prompts is a valid path, it is still beneficial to understand some of the basics of prompt engineering in order to effectively evaluate and adjust prompts in order to get to a working mentor model.
Keys to crafting a strong prompt
Prompting Best Practice | Less effective | More effective |
|---|---|---|
Be clear and specific: What exactly should the chatbot do? | "Help students with assignments." | "Guide students through the key steps of writing a lab report by asking them questions about their hypothesis, methods, and results. Provide tips for improving clarity and structure." |
Define the mentor’s persona: What tone and role should the chatbot adopt with students? | "Act like a tutor." | "You are a supportive, encouraging writing coach for first-year students. Use plain language and a warm tone. Always validate effort before offering suggestions." |
Include Examples: Show what good output looks like | “Help students with citations.” | "When a student asks how to cite a source in APA, respond like this: ‘Here’s an APA citation for a book: Author, A. A. (Year). Title of work. Publisher.’" |
Set Boundaries: What should the chatbot avoid or not do?
| "Don’t write essays for students." | "Support students in thinking through their ideas and organizing their responses, but always encourage them to write their own answers in their own words." |
Plan for Student Misuse or Misunderstanding: How might students game the system or get confused? | No guidance on misuse | "If a student asks you to ‘just give the answer,’ respond by saying: ‘Let’s work through the problem together. First, what do you already know about this topic?’" |
Embrace Iteration
Iterating is part of the process when it comes to writing good prompts. Your first try might not get exactly the response you want—but that doesn’t mean the idea is off. It usually just means you need to tweak the wording, add an example, or give clearer instructions. Each revision helps the AI better understand what you’re going for.
Think of prompt design like a conversation: you try something, see how the AI responds, then adjust. The more you play around with it, the better you get at getting the results you need. So don’t aim for perfection on the first go—just start, and refine as you go.
Resources for Prompt Creation
Prompt Library — More Useful Things: AI Resources A library of prompts for use in higher education, curated by Ethan Mollick and Lilach Mollick at Wharton.
Best practices for prompt engineering with the OpenAI API | OpenAI Help Center