Teaching · Practical guide
Use AI as a Questioning Tutor Without Giving Away the Thinking
An AI tutor can ask questions, but a long dialogue does not prove learning. The useful outcome is a learner who can explain and solve a new task with less help. This lesson demonstrates how to structure that interaction and how to recover when the assistant gives too much away.
Define what the tutor should do
A questioning tutor invites the learner to explain a claim, identify an assumption or take the next step. It should respond to the learner’s reasoning rather than delivering the entire solution immediately. In this lesson, “Socratic” refers to that questioning style; it is not a claim that a chatbot understands the learner or implements a validated tutoring method.
Set a narrow topic and provide a checked reference answer or source rule. For our exercise, the rule is percentage = subset divided by total, multiplied by 100, with a non-zero total. A bounded reference lets the teacher examine whether the assistant’s questions and hints stay within the intended content.
Agree on the support sequence before starting: ask for an attempt, ask a focused question, offer a small hint if needed, then provide an explanation when the learner remains stuck. Endless questioning can become frustrating. Direct explanation is sometimes appropriate; the important distinction is whether it supports the next independent attempt rather than replacing all of the learner’s work.
Choose questions that expose reasoning
A broad question such as “Do you understand?” invites a yes/no response. A more informative question asks, “Which number represents the whole group, and why?” It asks for evidence of a particular idea. If the learner supplies the wrong number, the next question should address that choice rather than introducing several new concepts at once.
Use a hint ladder. Level 1 points to the relevant idea without giving the number: identify the part and the whole. Level 2 supplies a partial structure: write part/whole before multiplying. Level 3 gives a worked step and asks the learner to explain it. These levels are an original planning device for this exercise, not a universal prescription.
Give the learner a way to ask for a direct explanation or pause. Do not frame requests for help as failure. Avoid prompts that ask the system to infer intelligence, motivation or a psychological state. The assistant can comment on the response it sees; it should not invent a profile of the person behind it.
Worked dialogue: correcting a denominator error
Task: 9 of 12 workshop places are booked. What percentage is booked? Learner: “I divide 12 by 9.” Tutor question: “Which quantity is the booked part, and which is all available places?” Learner: “9 booked and 12 available.” Tutor: “How would you write booked places as a share of all places?”
Learner: “9/12.” Tutor: “Good—the denominator now represents the whole. What is that ratio as a decimal?” Learner: “0.75.” Tutor: “How do you express 0.75 as a percentage?” Learner: “75%.” Tutor: “Explain in one sentence why 12, rather than 9, is the denominator.”
A suitable learner explanation is: “We compare the 9 booked places with all 12 available places, so 9/12 × 100 = 75%.” Notice that the feedback describes the corrected choice. It does not say the learner is naturally gifted, careless or incapable.
If the learner cannot form the ratio after two focused hints, the tutor can explain: “Use booked/available = 9/12. This equals 0.75, or 75%.” It should then offer a new task—such as 8 of 20 places booked—without providing the answer first. The new answer is 40%. That attempt gives more useful evidence than repeating the just-revealed answer.
Keep a human checkpoint
A model may misread a response, accept an incorrect answer or invent a confusing hint. Keep the checked source rule visible and pause if the dialogue contradicts it. Recalculate numerical examples independently. Asking the tutor whether it made a mistake can help locate the issue but is not a substitute for that independent check.
Use the conversation as low-stakes practice. Do not treat its praise, score or diagnosis as a formal assessment. For a real course, follow the teacher’s rules about permitted assistance and retain only the records required by policy. Practise without personal identifiers or protected assessment questions.
A useful exit checkpoint has three parts: solve a fresh example, explain the method and identify when it applies. For percentages, the learner should know that the denominator represents the reference total and cannot be zero. The checkpoint should be attempted before opening hints or model answers.
Practice: run and inspect a 20-minute tutoring exchange
Minutes 0–4: write a checked task and its answer. Use 6 completed exercises out of 15 assigned: 6/15 × 100 = 40%. Minutes 4–8: draft a likely incorrect learner response, such as “6%,” and prepare two focused questions that could expose the count-versus-percentage confusion.
Minutes 8–14: conduct a short interaction using the template, or write both roles on paper. Stop after six tutor turns. Minutes 14–20: attempt a fresh task and inspect whether each tutor question helped reveal a specific step. Record one question that was useful and one that gave too much away or was unclear.
Example questions are: “What does the number 6 count?” and “What fraction compares completed exercises with all assigned exercises?” A useful next hint is “Start with completed/assigned.” An unhelpful first hint is “The answer is 40%; can you repeat it?” because it removes the central calculation.
Fresh-task example: 7 completed out of 28 assigned gives 25%. An adequate explanation identifies 28 as the total and shows 7/28 × 100. Self-review: did the learner make an attempt? Did the support become more explicit only when needed? Was the final task completed and explained independently? These observations describe the exercise, not proof that a tutoring product improves learning.
Reusable prompt
Select and copy this template, or customise your own using the free builder.
Act as a questioning tutor for [topic] using only this checked rule: [rule]. Ask for my attempt first. Ask one focused question at a time about my reasoning. If I am stuck, offer a small hint, then a partial step, then a direct explanation if needed. Let me request an explanation or pause. Do not infer personal traits. After at most six tutor turns, give a fresh comparable task without its answer; reveal the explanation only after my attempt. Flag uncertainty and do not invent facts.Open the prompt builder →
Pause · Think · Check
Test your understanding
Choose one answer per question, then check your reasoning. This is practice; scores are not saved.