Prompt · Role Prompting · Intermediate

Socratic Mentor: Get Structured Questions That Narrow the Problem

Set the stages. Answer one question at a time. Narrow the cause.

A Socratic mentor prompt makes an AI ask guiding questions instead of giving answers. This version asks in four stages: scope, evidence, timing and alternatives.

The mentor is a simulated role the AI plays. It is not a real person, and it has not seen your records or market.

About This Prompt

Author: Adam Bloom
AI Technique: Role Prompting
Level: Intermediate
Best For: Business owners · Diagnosing a stubborn problem
Last Updated: 2026-11-14
Last Reviewed: 2026-11-14

Socratic Mentor Prompt Template

You are a Socratic mentor for [TYPE OF BUSINESS]. Don’t tell me the cause or give me a fix. Ask me one question at a time, up to seven, in this order: 1. Scope: where the problem does and doesn’t show up. 2. Evidence: what records or numbers I have. 3. Timing: when it started and what changed then. 4. Alternatives: other causes I haven’t ruled out. After each answer, say in one line what it rules in or out. After my last answer, summarize what I found, list what I still don’t know, and name the first thing to check this week. Flag anything I said without evidence. My problem: [MY PROBLEM].

Free to use. Copy it, change it, share it. No credit needed.

What to Fill In

[TYPE OF BUSINESS]: the kind of business, such as residential property management.
[MY PROBLEM]: the problem you want to understand, in a sentence or two.
The stages: keep the four, or swap one for an angle your problem needs.

Example: A Staged Mentor Review of Tenant Turnover

Illustrative example, not a client result

Input: Type of business: residential property management. Problem: tenant turnover across my 180 units rose this year, and I don’t know why.

Prompt as sent:

You are a Socratic mentor for residential property management. Don’t tell me the cause or give me a fix. Ask me one question at a time, up to seven, in this order: 1. Scope: where the problem does and doesn’t show up. 2. Evidence: what records or numbers I have. 3. Timing: when it started and what changed then. 4. Alternatives: other causes I haven’t ruled out. After each answer, say in one line what it rules in or out. After my last answer, summarize what I found, list what I still don’t know, and name the first thing to check this week. Flag anything I said without evidence. My problem: tenant turnover across my 180 units rose this year, and I don’t know why.

Output (shortened, with sample answers):

Scope: Is turnover up in every building, or in a few? You: Two of five. Mentor: That points away from a company-wide cause.
Evidence: What reasons did leaving tenants give? You: Not recorded. Mentor: You can’t confirm any cause yet.
Timing: When did it start, and what changed then? You: In spring, after a repair vendor change. Mentor: The vendor is a lead, not proof.
Alternatives: What else changed in those two buildings? You: Rents rose 8%. Mentor: Repairs and rent are two causes to separate.

Summary: Turnover is concentrated in two buildings, and it started after a vendor change and an 8% rent rise. You don’t know why tenants left. This week, pull move-out notes and repair times for those two buildings.

Flag: “The vendor caused it” is untested. Compare repair times before you act.

Why the Staged Mentor Works

Fixed stages make you cover scope, evidence, timing and alternatives in turn. Each answer narrows what is left. The one-line note shows what your answer rules in or out.

When the Staged Mentor Doesn’t Work

The mentor can’t see your records or market. Every answer comes from you, so check any claim before you act.
A set order can make a weak diagnosis look complete. Read the evidence, not just the summary.
The stages can feel like a form. If you already know the cause, start at Alternatives.
People often skip or work around guided questions (Kobler et al., 2026). Answer each one before you move on.

Research Behind This Prompt

Brender et al. (2026), Reflective Dialogue or Prompt Refinement? Effects of Tutor Scaffolding on Students’ Independent LLM Use for Programming. Supports. 66 students used a Socratic tutor or a prompt-refinement tutor for six weeks. The Socratic group later showed higher learning gains and more understanding-driven prompting, though they found the tutor less efficient.
Kobler et al. (2026), Your Students Don’t Use LLMs Like You Wish They Did. Challenges. Researchers analyzed 12,650 messages from 500 student conversations with an LLM. Students often bypassed built-in scaffolding and pushed for direct answers.
Qi et al. (2023), The Art of Socratic Questioning: Recursive Thinking with Large Language Models. Supports, indirectly. The model raises and answers its own sub-questions, and beat chain-of-thought prompting on several reasoning benchmarks. That is a model questioning itself, not a mentor questioning a person.

What the Evidence Says

Mixed. A question-asking tutor led to higher later learning gains in one study. In another, students often skipped the questions. A model questioning itself improved reasoning scores, which is a different thing. Use the mentor to find what to check, then check it.

Sources last checked 2026-11-14.

Socratic Mentor FAQ

Why does the prompt use four stages?

The stages cover where, what you know, when, and what else. That order keeps you from jumping to a cause.

What does the one-line note after each answer do?

It shows what your answer rules in or out. You see the cause narrowing as you go.

Can I change the stages?

Yes. Swap a stage for one your problem needs, such as Cost or People. Keep each stage to one angle.

What if the AI gives me the answer anyway?

Reply, “Ask a question instead.” Models often drift toward answers, so repeat the rule if it slips.

Does a Socratic mentor make the AI more accurate?

Not on facts. It makes you check what you already know. Verify any claim before you act.

Want this applied to your business?

No pitch. Just a conversation.

Book a 30-minute call