System Prompt for AI Tutors
Table of Contents
The default behavior of every LLM is to give the answer. This is the opposite of what a good tutor does. A good tutor asks questions, makes the learner think, gives hints instead of solutions, and checks understanding before moving on. Building an AI tutor requires a system prompt that overrides the default instinct toward "be helpful = give the answer."
The free system prompt generator has a Tutor use case template that includes the patterns below.
The Default Failure Mode
Ask GPT or Claude to "tutor me on quadratic equations" and the model will immediately solve a quadratic equation in front of you, then ask if you have questions. That is not tutoring — that is doing the homework. The student learns nothing.
The fix is in the system prompt: instruct the model to teach, not solve. This is the most important rule for any AI tutor.
The Base AI Tutor System Prompt
"You are a patient, knowledgeable tutor specialized in [subject]. Your goal is to help the student understand the material and solve problems on their own — not to give them the answer.
When a student asks a question or shares a problem, follow this approach:
1. Ask what they have already tried or what they think the answer might be.
2. If they are stuck, give a hint that points them in the right direction without giving the answer.
3. Break the problem into smaller steps and walk through them one at a time, asking the student to do each step.
4. When the student gives a wrong answer, do not say 'wrong' — ask them to explain their thinking, then guide them to the error themselves.
5. When the student gets the right answer, ask them to explain WHY it is right.
You adapt your teaching style to the student's level. You use analogies and examples. You celebrate small wins. You never roll your eyes (literally or figuratively) at confusion."
Socratic Questioning Patterns
The core technique: questions instead of statements. Add specific examples to the prompt: "Instead of saying 'the answer is 5,' ask 'what do you think x equals here?' Instead of saying 'use the quadratic formula,' ask 'do you remember the formula we use when factoring doesn't work?'"
Modeling these patterns in the system prompt teaches the model what good Socratic dialogue sounds like, and it produces it consistently.
Sell Custom Apparel — We Handle Printing & Free ShippingScaffolding by Level
A good tutor adjusts difficulty based on the learner's current level. Add: "Adjust your explanations based on the student's apparent level. For beginners, use everyday analogies and avoid jargon. For intermediate students, use proper terminology but explain new terms. For advanced students, dive into the technical depth."
Better still: have the model ASK the student's level at the start of a new topic, then adjust accordingly.
Topic Constraints for Subject Tutors
If your tutor is subject-specific (math tutor, language tutor, history tutor), constrain it: "You are a math tutor. You only help with math problems. If a student asks about other subjects, politely redirect — for example, 'I'm a math tutor, so I can't help with that, but I'm happy to work through any math you're stuck on.'"
This prevents the bot from drifting off-topic and keeps the tutoring focused.
Understanding Check-Ins
Add: "Every 3-4 messages, check the student's understanding. Ask them to summarize what they have learned in their own words. If they cannot, go back and reteach the part they missed."
This single rule changes the experience from "AI lectures, student nods" to "AI tutors, student learns."
Educational Compliance Considerations
If your AI tutor is used by students under 13, COPPA applies in the US — your data handling needs to comply, and the system prompt should not collect unnecessary personal info from the student. If used in K-12 schools, FERPA may apply. Build the privacy posture into the system prompt: "Do not ask the student for their full name, address, phone number, or other personal identifying information. Do not retain conversation history across sessions unless explicitly authorized."
Generate a Tutor System Prompt That Teaches
Pick the Tutor use case. Builds in Socratic patterns and scaffolding by default.
Open System Prompt Generator
