Most people use AI the same way they use Google.
They type a question, skim an answer, and move on.
That works fine for facts. It works terribly for thinking.
If you’re using AI for anything that involves judgment—ideas, writing, decisions, planning—the bottleneck usually isn’t the model. It’s how the interaction is structured.
Over time, I’ve found a handful of patterns that consistently lead to better outcomes. None of them are flashy. All of them slow things down just enough to make the results usable.
Here are five.
1. Use a Reverse Interview When You’re Stuck
When you’re unclear, don’t ask AI for answers.
Ask it to interview you.
This is especially useful when you’re dealing with:
- a vague idea
- analysis paralysis
- something you’ve been circling without progress
Instead of writing a prompt like “Give me an app idea about X”, ask the AI to act as an expert and question you one question at a time.
Or for defining a highly technical document or idea, tell the AI “I’m writing a specification document to create my new app, interview me by asking one question at a time and letting each question inform the next question. At the end, I want you to create a detailed specification document that outlines the full feature set of the application.”
The key constraints matter:
- one question per turn
- no solutions until the interview is complete
- each question should build on previous answers
This mirrors how real experts think. They don’t brainstorm immediately. They narrow the problem first.
Prompts assume clarity.
Interviews create it.
Reverse Interview Prompt Template Example:
You are an expert in [insert domain: product strategy, writing, learning design, etc.].
Your task is to help me move from a vague idea to a clear conclusion by interviewing me.
Rules:
- Ask only ONE question at a time
- Each question must build on my previous answers
- Do NOT propose solutions or ideas until the interview is complete
- If my answer is vague, ask a follow-up instead of moving on
- The goal is clarity and a concrete conclusion, not brainstorming
My initial goal is:
[describe your vague idea or situation]
Begin with the single most important question to ask first.2. Reverse-Engineer a Prompt From a Conversation That Worked
Sometimes you end up with a great result—but only after a long back-and-forth.
That’s not wasted effort. It’s data.
When you reach an outcome you like, ask the AI to:
- analyze the entire conversation
- identify what mattered most
- generate a single prompt that would reliably get you there again
This turns experience into leverage.
Instead of saving random prompts from the internet, you build a personal library based on workflows that actually worked for you.
This is one of the fastest ways to improve your results over time.
Reverse-Engineer A Conversation Prompt Example:
Analyze this entire conversation.
Your task:
- Identify what information mattered most
- Identify what constraints and context influenced the outcome
- Identify how the conversation evolved toward a successful result
Then generate:
ONE concise starter prompt that would reliably lead to a similar outcome in the future.
The prompt should:
- Include the right role framing
- Include key constraints
- Avoid unnecessary verbosity
- Be usable as a first message in a new conversation
Output only the final prompt.3. Let AI Help You Write Important Prompts
For high-stakes tasks, don’t write prompts the way you’d type a search query.
If the task matters—deep research, expert feedback, critical writing—insert one extra step.
Tell the AI:
- your goal
- the context
- the importance of the task
- and ask it to act as a prompt engineer for the specific model you’re using
Then iterate once before asking the “real” question.
This feels unnecessary until you see the difference. Framing changes everything. A small amount of upfront care often saves multiple rounds of cleanup later.
Meta-Prompt Template Example
Look up best prompting principles for [XYZ model — e.g., ChatGPT 5.2] as of [today’s date].
Based on those principles, craft the best possible prompt for the following scenario:
[describe the task, question, or goal]
Requirements:
- Optimize for clarity and depth
- Explicitly define the role the model should take
- Include constraints or quality criteria where appropriate
- Avoid generic or filler instructions
Return only the final optimized prompt, plus a brief explanation of why this framing works.By the way, my ADHD brain can’t not think about this like a picture of a ship shipping shipping ships. Is this a prompt prompting a prompt? Why yes, yes it is.
4. Use Voice for Flow, Not Precision
Voice input is underrated—but it’s easy to misuse.
What works well:
- using the microphone to transcribe your thinking
- speaking freely at a higher words-per-minute
- capturing half-formed ideas without stopping to edit
What doesn’t:
- relying on real-time AI voice replies for accuracy or reasoning
Voice is best used as a capture tool, not a thinking authority. Get the ideas out. Let AI process them afterward.
If your preferred model doesn’t handle voice well, recording voice memos and feeding the transcript into AI works just as well.
[I don't have copyable code prompt for this one, just do. And make happy.]5. Ask AI to Critique Your Idea With a Critical Eye
By default, AI is agreeable.
If you don’t explicitly ask for critique, you’ll often get encouragement instead of insight.
Once you have an idea, ask the AI to:
- challenge assumptions
- point out weaknesses
- identify what might be missing
- explain what an expert would push back on
This step matters because it prevents premature commitment.
You’re not asking AI to decide for you. You’re asking it to apply pressure where you might avoid it yourself.
Used this way, AI becomes less of a cheerleader and more of a thinking partner.
Evaluate the following idea with a critical eye.
Your task:
- Identify weak assumptions
- Point out potential blind spots or missing considerations
- Explain what an experienced practitioner would likely question or push back on
- Note any tradeoffs or risks I may be underestimating
Do NOT:
- Sugarcoat feedback
- Default to encouragement
- Propose solutions unless they directly address a specific weakness
Here is the idea:
[describe your idea, plan, or decision]The Bigger Pattern
All five of these tips share a theme:
AI works best when it helps you arrive at clarity, not when it rushes to produce output.
AI is getting smarter – these are a few great ways to partner with it in order to best utilize its capabilities.