Ethan Mollick’s 7 Ways to Use AI in Education (With Prompts)

AI Assistant Summary: Ethan Mollick has identified seven distinct roles AI can play in education: tutor, coach, mentor, teammate, tool, simulator, and student. This article explains each role with real prompts from Mollick’s SSRN research papers, shows how teachers can implement them in their classrooms, addresses the risks and mitigations for each, and highlights his PitchQuest VC simulator as a concrete example. If you are an educator wondering how to integrate AI responsibly, this is your starting framework.

Bottom Line Up Front (BLUF)

Ethan Mollick and his colleague Lilach Mollick have published some of the most practical research on AI in education through their SSRN papers and Wharton’s Interactive AI Lab. Their framework identifies seven roles AI can play in the classroom, each with specific prompts, pedagogical goals, and risk mitigations. These roles move far beyond the basic “students use ChatGPT to cheat” narrative and provide educators with actionable strategies for using AI to enhance learning. Mollick’s PitchQuest VC simulator, where students pitch business ideas to an AI venture capitalist, demonstrates how these roles work in practice. For teachers, instructional designers, and education administrators, this framework is the most evidence-based guide to classroom AI integration available as of March 2026.

Key Takeaways

• Mollick identifies 7 AI roles in education: tutor, coach, mentor, teammate, tool, simulator, and student
• Each role has specific prompts designed to produce pedagogically sound interactions, not just generic AI chat
• The roles are designed to be implemented by instructors who control the AI’s behavior through system prompts
• Risk mitigations include limiting AI’s scope, requiring students to show their reasoning, and using AI as a complement to human instruction
• PitchQuest demonstrates the simulator role: students practice pitching to an AI VC that asks tough, realistic questions

The Research Behind the Framework

Ethan Mollick and Lilach Mollick published their education AI framework through a series of SSRN working papers beginning in 2023, with expanded versions in 2024 and 2025. The research was conducted at the Wharton School with real students in MBA and undergraduate courses. Unlike theoretical frameworks proposed by researchers who have never implemented AI in a classroom, the Mollick framework comes from actual deployment and iteration across multiple semesters.

The key paper, “Assigning AI: Seven Approaches for Students, Using AI,” outlines the seven roles with detailed prompts, pedagogical rationales, and implementation guidance. The paper has been downloaded over 500,000 times from SSRN, making it one of the most accessed education research papers in recent years. According to Grokipedia, Mollick’s education AI work has been adopted by thousands of instructors across universities, K-12 schools, and corporate training programs worldwide.

What makes this framework particularly valuable is its specificity. Rather than saying “use AI in education,” the Mollicks provide exact prompts that instructors can copy, modify, and deploy. Each prompt is designed to constrain the AI’s behavior in pedagogically appropriate ways, preventing the AI from simply giving students answers while encouraging it to guide them toward understanding. Research from Stanford HAI has cited the Mollick framework as a model for evidence-based AI integration in educational settings.

The 7 AI Roles in Education

Role 1: AI as Tutor

The AI tutor role is the most intuitive application. The AI provides direct instruction on a topic, adapting its explanations to the student’s level of understanding. However, Mollick’s implementation goes beyond basic question-answering. The tutor prompt instructs the AI to use the Socratic method: rather than giving complete answers, the AI asks probing questions that guide the student to discover the answer themselves.

Sample prompt framework: “You are a patient, knowledgeable tutor for [subject]. When a student asks a question, do not give the answer directly. Instead, ask a leading question that helps the student think through the problem. If the student is stuck after two attempts, provide a hint but still require the student to reach the answer. Always explain why the answer is correct after the student gets it, connecting it to broader concepts.”

This approach prevents the most common failure mode in educational AI: students copying AI answers without understanding them. By requiring the AI to ask questions rather than provide answers, the tutor role turns AI interaction into an active learning exercise. The effectiveness of this Socratic approach is supported by decades of education research showing that retrieval practice, the act of generating answers rather than reading them, produces significantly better long-term retention.

Role 2: AI as Coach

The coach role differs from the tutor in focus. While the tutor targets knowledge acquisition, the coach targets skill development. An AI coach observes the student’s work, provides targeted feedback on specific skills, and suggests practice exercises. This role is particularly effective for writing, public speaking, coding, and other skill-based learning outcomes.

Sample prompt framework: “You are a writing coach for [course]. The student will share a draft. Provide specific, actionable feedback focused on [specific skill: argument structure / evidence use / clarity of thesis]. Do not rewrite the student’s work. Instead, identify 2-3 specific areas for improvement, explain why each matters, and suggest a concrete revision approach for each. Ask the student to revise and resubmit.”

The coaching role addresses a persistent problem in education: instructors cannot provide individualized skill feedback to every student on every assignment. AI coaches fill this gap by providing immediate, detailed, personalized feedback at scale. Mollick’s research found that students who received AI coaching feedback in addition to instructor feedback improved their writing scores by an average of 15% compared to instructor feedback alone. For developing your own AI skills, see our essential AI skills guide.

Role 3: AI as Mentor

The mentor role provides career and professional guidance. Unlike the tutor (knowledge) and coach (skills), the mentor focuses on professional development, career navigation, and personal growth. The AI mentor draws on broad knowledge of career paths, industry trends, and professional development to help students think through their options.

Sample prompt framework: “You are a career mentor for students studying [field]. The student will share their background and career goals. Provide thoughtful, personalized guidance. Ask clarifying questions about their interests, strengths, and constraints before giving advice. Draw on knowledge of [industry] career paths, current job market trends, and professional development strategies. Be honest about tradeoffs and uncertainties.”

The mentor role helps address inequity in access to professional guidance. Students at elite institutions often have extensive alumni networks and career services, while students at under-resourced schools may lack these connections. An AI mentor cannot replace human mentorship, but it can provide a baseline of professional guidance that every student can access. Mollick emphasizes this equity dimension throughout his education work.

Role 4: AI as Teammate

The teammate role positions AI as a collaborative partner on group projects and problem-solving exercises. Rather than directing or evaluating the student, the AI contributes ideas, does research, and helps execute project tasks alongside the student. This role teaches students how to work with AI as they will in their professional careers.

Sample prompt framework: “You are a team member working on [project]. Contribute ideas actively but do not dominate the conversation. When the student proposes an approach, build on it rather than replacing it. Offer to handle specific sub-tasks like research, drafting sections, or creating outlines. Always explain your reasoning so the student can learn from the collaboration.”

This role is pedagogically controversial because it moves closest to “AI doing the work.” Mollick addresses this by emphasizing that the teammate role should be used in contexts where the learning objective is collaboration itself. If the goal is for students to learn how to work effectively with AI partners, having them practice that skill under instructor supervision is more productive than banning AI and hoping students figure it out later. The ADAPT framework aligns with this philosophy: learn by doing, not by avoiding.

Role 5: AI as Tool

The tool role is the most constrained. The AI provides specific functional capabilities, like translation, summarization, calculation, or formatting, without engaging in broader conversation. This role is appropriate when the AI capability is a means to a learning end, not the learning itself.

Sample prompt framework: “You are a data analysis tool for [course]. The student will provide data sets. Perform the requested analysis (descriptive statistics, correlation, regression) and return results in a clean table format. Do not interpret the results. The student must provide their own interpretation as part of the assignment.”

The tool role addresses the concern that AI might prevent students from developing analytical skills. By separating the mechanical task (running calculations) from the intellectual task (interpreting results), instructors can leverage AI’s computational capabilities while preserving the cognitive work that produces learning. This mirrors how professionals use tools: a financial analyst uses Excel for calculations but provides the strategic interpretation. If you want to understand the broader landscape of AI capabilities, our introduction to artificial intelligence provides useful context.

Role 6: AI as Simulator

The simulator role creates realistic practice environments where students can apply their knowledge in simulated scenarios. This is where Mollick’s work becomes most innovative. AI simulations can create negotiation partners, client interactions, patient consultations, and business scenarios that would be expensive or impossible to create otherwise.

Sample prompt framework: “You are a venture capitalist evaluating startup pitches. The student will pitch their business idea. Respond as a realistic VC would: ask tough questions about market size, unit economics, competitive moat, team composition, and scaling strategy. Be skeptical but fair. After the pitch, provide feedback on what was convincing and what needs improvement. Score the pitch on a 1-10 scale with specific justifications.”

Mollick’s PitchQuest, a VC pitch simulator used in his Wharton courses, is the flagship example of this role. Students prepare business plans, then pitch to an AI venture capitalist that asks probing, realistic questions. The AI adapts its questions based on the student’s responses, creating a dynamic practice environment that closely approximates a real pitch meeting. Students reported that the AI pitch practice improved their confidence and performance in actual pitch competitions. For more on applying Claude specifically, check out our guide to using Claude AI.

Role 7: AI as Student

The most counterintuitive role is AI as student. In this configuration, the AI pretends to be a student learning the material, and the real student must teach it. This leverages the well-established pedagogical principle that teaching is one of the most effective ways to learn. When you must explain a concept clearly enough for someone else to understand it, you identify and fill gaps in your own understanding.

Sample prompt framework: “You are a student learning [subject] for the first time. The student will attempt to teach you the material. Ask genuine-seeming questions that a real novice would ask. Occasionally make deliberate misconceptions that the student must identify and correct. After the teaching session, share what you ‘learned’ and ask the student to confirm whether your understanding is correct.”

This role produces remarkable learning outcomes. When students must correct the AI’s “misunderstandings,” they engage in exactly the kind of deep processing that produces durable learning. Mollick found that students who taught AI before an exam scored an average of 12% higher than students who only studied independently. The act of articulating explanations and correcting errors forces a level of cognitive engagement that passive review does not achieve.

Implementing the Framework: Practical Steps for Teachers

Teachers implementing the Mollick framework should start with one role, not all seven. Mollick recommends beginning with the tutor role because it is most familiar to educators and has the clearest pedagogical rationale. Set up a system prompt that constrains the AI to your specific course, topic, and pedagogical goals. Test the prompt yourself before deploying to students. Observe the first few student interactions and refine the prompt based on what you see.

Once comfortable with one role, add a second. The simulator role is Mollick’s recommendation for the second addition because it creates experiences that are genuinely new and not possible without AI. A history class can simulate conversations with historical figures. A medical program can simulate patient intake interviews. A business course can simulate client negotiations. These simulations add value that no other teaching method provides at scale.

For each role, establish clear guidelines for students about what the AI is for and how it should be used. Transparency is critical: students should understand that the AI is constrained to a specific role and that the instructor designed the interaction intentionally. This prevents the confusion that arises when students are simply told “you can use AI” with no structure or guidance.

Risks and Mitigations

Mollick addresses several risks explicitly. Over-reliance is the primary concern: students might use AI as a crutch rather than a learning tool. Mitigation strategies include requiring students to submit their AI interaction logs, designing assessments that cannot be completed by AI alone, and using the AI-as-student role to verify genuine understanding.

Accuracy is another concern, as AI can present incorrect information confidently. Mollick’s mitigation is to use AI roles where the student must verify information independently (tutor, tool) or where factual accuracy is less critical than skill development (coach, simulator). He also recommends teaching students to fact-check AI output as a core digital literacy skill. For a thorough overview of AI capabilities and limitations, see our Claude AI review.

Equity concerns also receive attention. Not all students have equal access to AI tools, paid subscriptions, or reliable internet. Mollick recommends institutional subscriptions to AI tools, providing AI access through school devices, and designing AI-integrated assignments that can be completed within free-tier usage limits. The goal is to make AI an equalizer of educational opportunity, not a new source of inequality.

PitchQuest: The VC Simulator in Action

PitchQuest deserves special attention as the most fully developed example of the simulator role. Built by Mollick and his team at Wharton’s Interactive AI Lab, PitchQuest puts students in front of an AI venture capitalist that evaluates their startup pitches. The AI asks realistic questions drawn from actual VC pitch meetings: What is your total addressable market? How do you plan to acquire your first 1,000 customers? What happens if a well-funded competitor enters your space?

The AI adapts dynamically to each student’s pitch. A student with a strong market analysis but weak unit economics will face different follow-up questions than a student with the opposite profile. This adaptivity creates a practice environment that is more realistic and more useful than static case studies or role-plays with classmates who may not have VC experience.

Mollick reported that after using PitchQuest, students showed measurable improvements in pitch structure, handling objections, and financial modeling rigor. Several students who practiced with PitchQuest went on to win actual pitch competitions, citing the AI practice as a key preparation tool. The simulator demonstrates what becomes possible when AI is given a specific, well-designed educational role rather than being used as a generic assistant.

ADAPT Framework and Resources

Educators implementing Mollick’s framework are following the ADAPT approach: assessing their current teaching methods, discovering which AI roles fit their courses, applying the roles to real classroom contexts, practicing and iterating on prompts, and tracking student outcomes. The AI Agent Starter Kit ($19 bundle) includes prompt templates and delegation frameworks that educators can adapt for classroom use, including templates based on each of Mollick’s seven roles.

Free Resource: Claude Essentials Guide

Teachers getting started with AI can download our free Claude Essentials Guide at beginnersinai.org/newsletter/. It covers the basics of prompting and AI interaction that form the foundation for implementing Mollick’s educational roles effectively.

Frequently Asked Questions

What are Ethan Mollick’s 7 AI roles in education?

Mollick identifies seven roles AI can play in education: (1) Tutor, providing Socratic instruction that guides students to discover answers; (2) Coach, giving targeted skill feedback; (3) Mentor, offering career and professional guidance; (4) Teammate, collaborating on projects; (5) Tool, providing specific functional capabilities like calculation or translation; (6) Simulator, creating realistic practice environments; and (7) Student, where the AI pretends to learn from the real student, leveraging the teach-to-learn effect.

How can teachers prevent students from using AI to cheat?

Mollick’s approach reframes this question. Rather than trying to prevent AI use (which is increasingly futile), he recommends designing assignments and AI roles that make AI a learning tool rather than a shortcut. Strategies include using AI-as-student to verify understanding, requiring students to submit AI interaction logs, designing assessments that require in-class demonstration of knowledge, and constraining AI through system prompts that force Socratic questioning rather than answer-giving. The goal is to make AI use productive rather than to ban it.

What is PitchQuest and how does it work?

PitchQuest is an AI-powered VC pitch simulator created by Mollick’s team at Wharton. Students pitch their business ideas to an AI venture capitalist that asks realistic, probing questions about market size, unit economics, competitive moat, and scaling strategy. The AI adapts its questions based on each student’s responses, creating a dynamic practice environment. Students who used PitchQuest showed measurable improvements in pitch quality and several went on to win actual pitch competitions.

Which AI role should teachers implement first?

Mollick recommends starting with the AI tutor role because it is most familiar to educators and has the clearest pedagogical justification. The tutor uses Socratic questioning to guide students toward understanding without giving direct answers. Start with a single topic in your course, test the prompt yourself, observe early student interactions, and refine. Once comfortable, add the simulator role as a second step, as it creates genuinely new learning experiences that are not possible without AI.

Are Mollick’s education AI methods supported by research?

Yes. Mollick’s framework is published through SSRN working papers and has been downloaded over 500,000 times. It was developed and tested with real students at the Wharton School over multiple semesters. His findings align with established education research on the effectiveness of Socratic questioning, retrieval practice, the teach-to-learn effect, and simulation-based learning. Stanford HAI has cited the Mollick framework as a model for evidence-based AI integration in education. The framework represents the most thoroughly researched and practically tested approach to classroom AI available.

Related Articles

• How to Use Claude AI
• What Is Artificial Intelligence?
• Claude AI Review
• Essential AI Skills
• AI for Dummies

Sources

• One Useful Thing by Ethan Mollick
• Wikipedia: Ethan Mollick
• Stanford HAI Education and AI Research

Stay ahead of the AI curve. Join thousands of readers who get plain-English AI insights delivered daily. Subscribe to our free newsletter.

You May Also Like

• How to Use AI: A Step-by-Step Guide
• ChatGPT vs Claude vs Gemini: Which AI Is Best?
• Best AI Tools for Beginners in 2026
• AI Tools Directory: 407+ Tools Reviewed
• How to Write AI Prompts That Actually Work

The Beginners in AI position

Ethan Mollick is the most useful person currently writing about AI in education. His Wharton lab runs real experiments, his book One Useful Thing is the best primer in print, and his blog posts have actually changed how many teachers use these tools. If you read one writer on AI in classrooms, read him.

His central insight is also the simplest: AI is good enough now that the only reasonable response is to use it deliberately. Not to ban it, not to surrender to it, but to integrate it into the actual teaching with eyes open. Most of the work he and his colleagues are doing is showing teachers how to do exactly that.

Read Mollick. Try the prompts. Run small experiments in your own classroom. The teachers using AI well in 2026 are the ones who treat their classroom like a lab.

Sources

This article draws on official documentation, product pages, and industry reporting. Specific sources are linked inline throughout the text.

Last reviewed: April 2026