The Socratic Prompt: How to Make AI Think Like Your Best Consultant
Most AI prompts are shopping lists.
“Write me a LinkedIn post about B2B sales.” “Draft an email to this client.” “Summarise this document.”
Shopping lists get shopping-list outputs. Generic. Shallow. The kind of thing you could have produced yourself in five minutes and wouldn’t be proud of.
There’s a better way. It’s called Socratic prompting. It’s the difference between ordering a cheap takeaway and asking a proper consultant to think through your problem.
Here’s the trick. Before you ask for output, you ask for thinking. You force the AI to reason step-by-step, the way a senior consultant would, before producing the answer.
The output gets better. Immediately. Dramatically. And it costs you nothing extra.
Let me show you how.
Task prompts versus thinking prompts
A task prompt gives an instruction. “Write a product description for my widget.”
A thinking prompt gives a process. “Before writing the product description, list the three main use cases for this widget, identify the ideal customer, and name the one objection a buyer is most likely to raise. Then write the description.”
Same tool. Same model. Completely different output.
The task prompt gets you generic marketing waffle. The thinking prompt gets you something that sounds like it was written by someone who actually understands your product.
The magic is in the “before.” You’ve forced the AI to do the analysis it would normally skip. The final output is informed by that analysis. Everything gets sharper.
This is how consultants work. They don’t start drafting the slide deck on day one. They ask questions first. Who is this for? What are they trying to achieve? What’s in their way? What’s worked before?
Those questions shape the output. Without them, you get a template. With them, you get insight.
Why shallow prompts produce shallow output
AI models work by pattern matching. You give them an input. They produce the most likely output based on everything they’ve read.
When the input is “write a LinkedIn post about sales,” the most likely output is a generic sales post. There are millions of those. The AI averages them.
When the input is specific — with constraints, context, and thinking — the output can’t be the average. It has to fit your constraints. It has to reflect your context. It has to pass through your thinking before it writes.
Specificity forces quality. Vagueness invites average.
Most owners complain that AI produces boring writing. They’re right. But the AI isn’t being boring. The prompt is being vague. Fix the prompt, and the quality changes overnight.
Three Socratic templates
Here are three templates you can steal today. Each one follows the same pattern: question first, output second.
Template one: strategy
Use this for anything where you need a strategic opinion.
I'm working on [problem/decision].
Before recommending an approach, answer these questions:
1. What are the three most common ways businesses solve this?
2. What are the weaknesses of each approach?
3. What information would a good consultant need before giving advice?
4. Based on what you know, what would you ask me next?
Then, given your analysis, recommend an approach.The AI has to think. It names the options. It critiques them. It identifies what it doesn’t know. Only then does it recommend.
Run this instead of “what should I do about my pricing?” and watch the quality jump.
Template two: content
Use this for writing tasks.
I'm writing a [blog post / newsletter / LinkedIn post] about [topic].
Audience: [specific description — who they are, what they know, what they care about]
Before writing, answer:
1. What are the three most common takes on this topic already published?
2. What's a non-obvious angle most people miss?
3. What's the one thing my audience needs to understand or do after reading?
Now write the post, using the non-obvious angle, ending with a clear next action.This forces the AI to scan the obvious and pick something sharper. It also sets a purpose: not just to inform, but to change behaviour.
Template three: problem-solving
Use this for operational or technical problems.
I have a problem: [describe it in plain terms]
Before suggesting a solution, work through:
1. What are the possible root causes, ranked by likelihood?
2. What questions would help me diagnose the real cause?
3. What are the costs of misdiagnosing this?
4. What's the lowest-risk first step to test the most likely cause?
Then, recommend the first step.Instead of jumping to a solution, the AI has to diagnose. You get a tested recommendation instead of a guess.
How to build a prompt library
Good prompts are assets. Treat them that way.
Start a shared document. Google Doc, Notion, a plain text file in your vault. Call it “Prompts.”
Every time you write a prompt that produces a great output, save it there. Note:
What you were trying to do
The full prompt you used
What worked about the result
Anything you’d tweak next time
After a month, you’ll have twenty or thirty reusable prompts. After six months, you’ll have a hundred.
Organise by use case. Sales. Content. Strategy. Research. Admin.
When a team member needs to do a similar task, they don’t start from scratch. They grab the prompt, adjust the specifics, run it.
This is how small teams outperform big ones. Shared assets. Shared thinking. Compounding value.
The CLAUDE.md pattern
Here’s a next-level move. Most teams don’t do this. The 11% do.
Create a single file that describes your business, your voice, your standards, your context. Save it in your prompt library. Every time you use AI for anything non-trivial, paste the context at the top.
For me, that file is called CLAUDE.md. It lives in my T40 OS vault. It covers:
Who I am and what my business does
My voice (British English, Hemingway Grade 4, no AI slop)
My target customers
My current priorities for the quarter
My key people and active projects
My values and what I won’t compromise on
Every significant AI conversation starts with that context. The AI knows who it’s helping. It knows the rules. It knows the goal.
Outputs get sharper. Errors drop. Consistency goes up.
This is the shared team brain the 11% build. You don’t need fancy software. You need a text file, a commitment to keep it updated, and the discipline to use it.
What to avoid
Three common mistakes when you start using Socratic prompts.
Asking too many questions. Three to five guiding questions is the sweet spot. Ten is overkill. The AI starts producing a checklist instead of reasoning.
Leading the witness. Don’t ask questions that telegraph the answer you want. “What are the weaknesses of the competitor’s product?” is fine. “Why is the competitor’s product terrible?” is a loaded question. The AI will produce whatever confirms your bias.
Skipping the thinking output. Let the AI show its work. Don’t say “think about this but don’t tell me your thinking.” The act of writing the analysis changes the final output. You benefit from reading it too — sometimes the analysis is the answer.
Why this works
Two reasons.
First, AI models produce better output when they’re forced to reason step-by-step. This is well-documented in research. Chain-of-thought prompting improves accuracy on complex tasks by thirty to fifty percent in some benchmarks.
Second, the process of writing a Socratic prompt forces you to think clearly. Half the benefit is in the prompt construction. You clarify what you actually want. The AI is then more likely to deliver it.
It’s a two-way upgrade. The AI reasons better. You brief better. The gap between mediocre and excellent closes.
The practical bit for this week
Next time you sit down with an AI tool for a non-trivial task, pause.
Before you type the task, type three guiding questions.
Example. You want a value proposition for a new service. Instead of “write me a value proposition,” try:
Before writing a value proposition for [service], answer:
1. What does a strong value proposition always contain?
2. Who is the ideal customer for this service, and what are their top three pains?
3. What are the two most common objections someone might raise?
Now write three value propositions. Each should be one sentence.Compare the output to what you’d have got from a plain task prompt.
Do this five times this week. After five, you’ll never write another shopping-list prompt.
The big shift
The difference between an AI user and an AI operator is the prompt.
Amateur users give AI tasks. Operators give AI thinking frameworks.
Amateurs get drafts. Operators get insight.
Amateurs cancel their subscription after three months because the outputs feel shallow. Operators compound their skill every month because their prompt library keeps getting sharper.
It’s the same tool. The same £20 a month. Completely different businesses built on top of it.
The bottom line
Better prompts produce consultant-level thinking. Not faster drafting. Actual analysis. Actual reasoning.
Ask guiding questions first. Request the output second. Save what works. Share it with your team. Maintain a shared context document.
That’s the Socratic prompt. Simple. Free. Transformational.
Use it today. Your next AI output will be better than your last one. Keep using it, and your AI stops being a novelty and starts being a consultant.
Sources:
Industry research on chain-of-thought and Socratic prompting — forces step-by-step reasoning, significant quality improvement
Team best practice — shared prompt library drives adoption and consistency
CLAUDE.md pattern — shared context file as a team brain
Practical action for this week: Before your next five AI requests, write three guiding questions first. Compare outputs to your usual prompts. Save the best results to a prompts document.
Next week: why one person should own your AI automation.