Built to be screenshotted and used Monday morning.
The mental model, the 5 techniques, and the prompt templates from my AI Summit talk.
Before any technique works, you need this in your head. It's the difference between using AI like a Google search and using it like an employee.
Imagine you read 100,000 books — but can't remember a single sentence from any of them.
What you DO have is a deep gut feeling for what word comes next in any sentence. That gut feeling is the LLM.
It does that one word at a time, a thousand times in a row, and you get an essay. No database lookup. No live Wikipedia. Just prediction based on patterns it absorbed during training.
This is also why it hallucinates — it's not LOOKING anything up. It's predicting what sounds right.
Vague words → average prediction. Specific words → specific prediction. That's prompting in one line.
Meta-prompting is the next level up: instead of trying to write the perfect prompt, you use the AI to help you build it. You make the AI build its own steering wheel.
Each one is a technique, not a use case. Apply any of them to any problem — writing, research, code, strategy. The order below is roughly how you'd use them on a single big task.
Most people tell AI how to do the work. The pros tell it what the finished work looks like and let it figure out the steps. The AI is better at planning than you think.
I want a landing page for [product] that converts cold traffic from Meta ads at > 3%. The reader is [audience]. Voice should feel like [reference brand]. Don't give me a checklist. Give me the finished page, then explain the choices.
You know more than you can articulate. Flip the question direction — let the AI ask the questions and pull the context out of you. Output quality goes up 10x with the same starting effort.
Before you write anything, ask me the 10 most important questions you'd need answered to do this brilliantly. Ask them one at a time. Wait for my answer. Don't proceed until you've got what you need.
Don't write the prompt yourself. Have the AI write the prompt, then have it grade its own work. AI is a better critic of its own output than you are.
Write me the best possible prompt to accomplish [goal].
Then, in a second pass: review the prompt you just wrote. What's missing? What would a senior prompt engineer cut? What would a skeptical CMO add? Rewrite it incorporating your own feedback.
The killshot. When the output is mediocre, most people ask for a better output. Instead: ask what's wrong with the PROMPT that produced it. Then update the prompt and re-run. You're compounding an asset every time.
This output isn't good enough. Don't rewrite the answer. Tell me what's missing from my PROMPT that caused this mediocre answer. Then give me the upgraded prompt. I'll re-run it.
Same prompt. Two models. Different universes. Run your prompt in Claude AND ChatGPT AND Gemini side-by-side. Pick the winner. Then hand the loser's output to the winner and ask it to improve. This is how you stop trusting one brain and start running an internal council.
[Run the same prompt in Claude, ChatGPT, and Gemini.] Then paste all three outputs into one of them and say: I ran the same prompt in three models. Here are the three outputs. Tell me which is strongest and why. Then write a synthesis that takes the best of each. Be honest about which model won and where.
Copy-paste these directly into Claude or ChatGPT and replace the [brackets]. They use the 5 techniques baked in.
For writing a high-conversion landing page from scratch.
You are the world's most successful direct-response copywriter. You've written landing pages that have generated over $1B in revenue. I want you to write the highest-converting landing page possible for: - Product: [describe] - Audience: [who they are, what they want, what they fear] - Price point: [$] - Voice reference: [brand or writer whose tone fits] Before writing anything, ask me the 7 questions you need answered to do this brilliantly. Ask one at a time. Once you have my answers, draft the page. Then critique your own draft — what's weak, what's generic, what would a skeptical buyer skip past? Rewrite the page incorporating that critique. Output the final version only. No preamble.
For getting a sharp second opinion on any business decision.
You are my brutally honest advisor — the version of a senior partner who tells me the truth instead of what I want to hear. I'm facing this decision: [describe situation] My current lean: [what I'm thinking of doing] Before you weigh in, ask me 5 questions that would change your advice if the answers went one way or the other. Then give me your real take. Tell me what I'm probably wrong about. Tell me what I'm not seeing. End with: "If I were you, I would ___." No diplomatic hedging. I'm here for the real answer.
For when you have a vague idea and need to turn it into a sharp plan.
I have a rough idea: [1-2 sentence description]
I want to turn this into a real, executable plan. But I don't know what I don't know.
Your job: ask me 15 questions, one at a time, that will surface the assumptions, gaps, and decisions I haven't thought about. After each answer, briefly tell me what your follow-up question is going to probe — so I can see where you're steering me.
At the end, synthesize my answers into:
1. The sharpest version of my idea
2. The biggest risk
3. The first 3 things I should do this week
Run this in 2 or 3 models. Then paste all outputs back into one to crown a winner.
Task: [your task] Constraints: - [constraint 1] - [constraint 2] Audience: [who this is for] Deliver your best possible attempt. After your draft, score yourself 1-10 on these dimensions: - Specificity - Originality - Usefulness - Voice Then in one sentence, tell me what you'd do differently if you had a second pass.