Socrates Was the World's First — and Best — Prompt Engineer

A developer argues that Socrates was the first and best prompt engineer, drawing parallels between Socratic questioning and modern LLM prompting techniques. The post maps six Socratic question types to contemporary prompting methods, such as chain-of-thought and multi-agent debate, and provides ready-to-use prompt templates. It suggests that advanced prompting techniques are modern repackagings of Socratic methods.

Li Fei-Fei said something that keeps showing up in AI circles: "If Socrates were alive today, he would be the best prompter. Prompting is fundamentally the art of asking questions to pursue truth." Most people share it and move on. I want to actually prove it — not as a fun philosophical exercise, but as a practical toolkit you can start using today. Socrates 470–399 BC never gave a single answer in his dialogues. He designed questions that forced people to expose their own contradictions, revise their assumptions, and arrive at clearer thinking on their own. Sound familiar? That's exactly what the best LLM prompts do. This article does two things: Socrates never said "justice means X." He would set the scene "we're talking about justice in a city-state" , let the other person offer a definition, then systematically expose its contradictions — until the person revised their own answer. The prompt engineering translation: In Meno , he teaches a slave boy geometry without ever giving a correct answer. He just asks questions until the boy figures it out himself. This is maieutics midwifery — the idea that knowledge is already inside the person/model; the prompt's job is to deliver it. R.W. Paul documented six types of Socratic questioning. Each one maps cleanly to a modern prompting technique: | Socratic Question Type | Modern Prompt Equivalent | |---|---| | Clarify concepts | Rephrase & Respond | | Expose assumptions | Self-Consistency / Adversarial check | | Request evidence | Chain-of-Thought step-by-step reasoning | | Shift perspective | Multi-agent debate | | Trace consequences | Tree of Thoughts | | Question the question | Self-Ask / Maieutic prompting | Every "advanced prompting technique" you read about in 2026 blog posts is a modern repackaging of something Socrates was doing 2,400 years ago. Here's the difference: ❌ Generic prompt : "Help me write a marketing plan." → Output: confident-sounding generic advice ✅ Socratic-style structured prompt : "You are a B2B SaaS growth expert. For a startup in months 0–6 post-launch, propose step-by-step acquisition strategies. For each step, state the underlying assumption and how you would validate it." → Output: structured, falsifiable, expert-level strategy Socrates knew this intuitively: the question isn't asking for an answer — it's structuring the problem so the answer becomes visible. This is the entire discipline of prompt engineering. LLMs have a core problem: they produce fluent BS. Socrates faced the same problem with Sophists — people who argued eloquently for false things. His answer was elenchus refutation : systematically testing a claim with counterexamples until contradictions surface. "You said courage is fearlessness. Does that mean a soldier who knows the battle is lost and fights anyway be brave, or foolish?" Modern equivalent: "Before responding, identify three assumptions in the user's request, two potential counterexamples to your answer, and one way your reasoning could be wrong." That's Socratic elenchus in a prompt. It forces the model to self-check — exactly the way Socrates forced his interlocutors. He wasn't just first though he was early . He was best because he simultaneously achieved three things that most prompt engineers only achieve one or two of: Below are the six Socratic question types, mapped to ready-to-copy prompt templates. Use them as-is or adapt them to your domain. Socratic purpose : Make sure you understand what is actually being said before reacting. When to use : When the user's request is vague, when you want to force precise thinking, or as a first step in any analysis. Prompt template : Before answering, do the following: 1. Rephrase the user's request in your own words, using different wording. If your rephrasing diverges significantly from the original, flag the gap. 2. Identify what type of answer is expected a definition, a procedure, a judgment, a prediction, etc. . 3. If the request contains any ambiguous terms, state your interpretation and proceed with that interpretation explicitly noted. Do not answer the question yet — only clarify it. Socratic purpose : Surface what is being taken for granted that might not hold. When to use : Before making strong claims, when analyzing a decision, or when evaluating someone else's argument. Prompt template : Before giving your answer, systematically examine the assumptions embedded in the question or claim: 1. List every assumption the question or argument relies on. Be explicit — don't assume the user or I already knows these. 2. For each assumption, state whether it is: - Well-supported explicitly stated in the prompt - Reasonably inferred - Your own unstated interpolation 3. Pick the two assumptions that feel most fragile or least examined, and explain why they might not hold. 4. If removing or reversing one of these assumptions would significantly change your answer, note that explicitly. Then proceed with your answer, incorporating this assumption audit. Socratic purpose : Demand reasons, not just conclusions. When to use : When asked to evaluate, compare, recommend, or judge. Almost any substantive question. Prompt template : You are about to give an answer. Before you do, walk through the reasoning chain step by step. For each step in your reasoning: - State the claim being made - Give the specific evidence or principle that supports it - Note whether that evidence is empirical, logical, conventional, or an assumption If you encounter a step where evidence is weak or missing, say so explicitly rather than filling the gap with plausibility-sounding language. Your final answer should be traceable back to these steps — a reader should be able to see why you concluded what you concluded, not just what you concluded. Structure your response as: Reasoning Chain step-by-step Conclusion answer Evidence Gaps anything you couldn't fully support Socratic purpose : Break the single-voice bias by generating opposing positions. When to use : Strategic decisions, evaluating trade-offs, when you suspect groupthink, before any irreversible action. Prompt template : You will generate three distinct perspectives on the question/task at hand. These are not positions you personally endorse — they are rigorous arguments for different positions. Perspective A Advocate : Make the strongest possible case for the most obvious or popular position. Cite real-world analogies and concrete examples. Perspective B Devil's Advocate : Make the strongest possible case against the popular position. Identify the failure modes, second-order effects, and uncomfortable truths the popular view glosses over. Perspective C Third Way : Identify what both A and B are missing or misframing. Propose a synthesis or reframing that neither side adequately addressed. After presenting all three perspectives, conclude with: - Which perspective has the strongest structural argument not the most compelling narrative - Where all three perspectives share a hidden assumption that might invalidate all of them - Your own considered judgment, with explicit reasoning Socratic purpose : Follow the implications of a position or decision to see where it actually leads. When to use : Decision-making, evaluating plans, policy analysis, any "should we do X?" question. Prompt template : You are evaluating a claim / a decision / a proposal . Map out the consequences using a tree structure: Immediate consequences : What happens in the first day / week / month if this is adopted? Secondary consequences : What changes as a result of the immediate effects? Consider: who benefits, who loses, what new problems are created? Tertiary consequences : What are the second-order effects at 1 year? At 5 years? Failure mode : If this goes wrong, what is the worst-case scenario, and how likely is it? Success condition : Define what "success" would actually look like. What metrics would tell you this is working? Pre-mortem : Assume the decision was made and it failed badly. Now work backward — what specific decisions or低估ed factors caused the failure? End with a summary: given this full consequence tree, what is the expected value of this decision, and what would change your recommendation? Socratic purpose : Step back and ask whether the question itself is the right question. When to use : When stuck, when the problem feels circular, when solutions keep failing, when you suspect the real issue hasn't been named. Prompt template : Before attempting to answer the user's question, conduct a meta-level examination: 1. Reformulate : The user asked " restate the question ". Is this the actual question they need answered, or is it a symptom of a deeper question? 2. Identify the real question : If the user is asking "how do we X?", they may actually need to know whether they should X at all. What is the underlying question that a good answer would address? 3. Test the framing : Is the question framed in a way that makes a good answer possible? If the question contains a false assumption, say so and explain why. 4. Check for XY problems : Is the user describing their attempted solution Y instead of their actual problem X ? If so, name X explicitly. 5. Propose the best version of this question : If you were to ask the most precise, most answerable version of what the user seems to need, what would it be? Then answer the reformulated question — and note explicitly where your answer differs from what you would have said to the original question, and why. For high-stakes or complex queries, run the six questions in sequence: You are conducting a Socratic analysis of the following question/topic: User's input Follow this exact sequence: STEP 1 — CLARIFY: Rephrase and identify ambiguities STEP 2 — ASSUMPTIONS: List and stress-test hidden assumptions STEP 3 — EVIDENCE: Trace your reasoning with explicit justification STEP 4 — PERSPECTIVES: Generate three opposing viewpoints rigorously STEP 5 — CONSEQUENCES: Map immediate, secondary, and tertiary outcomes + pre-mortem STEP 6 — META: Step back — is this the right question? Reformulate if needed. Then provide your final synthesized answer, incorporating everything discovered in steps 1–6. This is the complete Socratic protocol. Run it when: A necessary caveat: this analogy has a clear scope. Socrates was working with people — beings who feel pain, hold stubborn beliefs, grow over time, and care about truth for its own sake. His goal was arete excellence of character and anamnesis recollection of knowledge the soul already has . Prompt engineering is working with statistical models — systems that optimize for plausible text, not truth for its own sake. The goal is reliable, useful output . The analogy holds perfectly at one layer: "structured questioning controls and improves reasoning." Beyond that layer, pushing the metaphor produces warm fuzzy feelings instead of useful insights. Socrates was a philosopher shaping souls. Prompt engineering is interface design for prediction engines. Both ask great questions. Only one of them can feel embarrassed. | Socratic Question | Prompt Template | Primary Use | |---|---|---| | Clarify | Rephrase & Respond | Vague requests, precision thinking | | Expose Assumptions | Self-Consistency Check | Strong claims, evaluation, analysis | | Request Evidence | Chain-of-Thought | Justification, reasoning, judgment | | Shift Perspective | Multi-Agent Debate | Decisions, trade-offs, groupthink | | Trace Consequences | Tree of Thoughts | Planning, policy, risk analysis | | Question the Question | Meta-Reasoning | Stuck problems, reformulation | The next time someone says "Socrates would be a great prompter," you can do more than nod along. You can pull up these templates and show them exactly what that means in practice. If you found this useful, the previous article in this series covers the full AI-assisted programming stack — from Transformers through Agentic AI.