Framework
Stop Prompting. Start Operating.
By MrPrompts
Most people type a question into AI and hope for something useful. Sometimes it works. Usually the output is generic.
The people I see getting consistently good results are not writing better questions. They are giving the AI short, specific instructions that tell it how to think about the task. That is what this page is about.
What this is and how to use it
Below are 15 instructions you can type directly into any AI tool, whether that is Claude, ChatGPT, Gemini, or Perplexity. No special software. You just type the instruction before your request.
Each instruction tells the AI how to approach your task. You are not hoping for a good answer. You are specifying what kind of output you need, how it should be structured, and where to focus.
A quick example. Say you need help with a marketing plan. Instead of typing "give me a marketing plan," you type:
DefineOutcome: A 30-day marketing plan that generates 50 qualified leads AddContext: We are a 20-person consulting firm targeting CFOs ForceAction: Give me the exact steps to execute this week CheckGaps: What am I missing that could make this fail?
Four short instructions. Now the AI knows what success looks like, who you are, that you want steps (not ideas), and that it should find the holes in its own thinking. The difference in output quality is significant.
You can use one instruction or stack several. Start with one. Add more as you get comfortable. They work on their own, but they compound when combined. I typically use two or three together and that covers most of what I need.
The 15 Instructions
Copy any of these and type it into your AI tool before your request. The examples show you exactly what to write.
DefineOutcome
Tell the AI what a good answer actually looks like. Without this, it guesses.
DefineOutcome: A 90-day onboarding plan that gets new hires to full productivity, with weekly milestones and clear ownership for each task.
AddContext
Give it the background it needs so you get a relevant answer instead of a generic one.
AddContext: We are a 200-person B2B SaaS company. Our sales cycle is 45 days. Average deal size is $35K. We sell to VP-level buyers in financial services.
AssignRole
Put the AI in a specific mindset. A CFO thinks differently than a marketing lead, and the output reflects that.
AssignRole: You are a CFO preparing a board presentation on Q1 results. Your audience is non-operational board members who care about trends, not details.
SetConstraints
Limit scope, length, or style. Tighter constraints produce tighter answers.
SetConstraints: Under 200 words. No bullet points. Written for a CEO who has 90 seconds to read this before a meeting.
SpecifyFormat
Tell it exactly how to structure the output so you can use it right away.
SpecifyFormat: Table with columns: Initiative, Owner, Deadline, Status, Risk Level. Sort by deadline.
ForceAction
This is the one I use most. It stops the AI from giving you ideas and makes it give you steps.
ForceAction: End with exactly 3 actions I can take this week. Each action must include who does it, what they do, and by when.
AskUserQuestion
Have the AI ask you questions before it answers. This catches bad assumptions early.
AskUserQuestion: Before answering, ask me 3 questions that would make your response significantly more useful.
ChallengeThinking
Push past the obvious. The AI defaults to safe, conventional answers unless you tell it not to.
ChallengeThinking: What is the contrarian view here? What would a skeptic say? Where is the conventional wisdom wrong?
CheckGaps
Ask the AI to audit its own answer. It is surprisingly good at finding what it missed.
CheckGaps: Review your answer. What did you leave out? What assumptions did you make? What risks did you not address?
ImproveOutput
Refine what you already have instead of starting from scratch.
ImproveOutput: Make this sharper. Remove anything that does not directly support the main argument. Tighten the language.
Simplify
Strip out complexity. I use this whenever output comes back full of jargon.
Simplify: Rewrite this so a smart 14-year-old could understand it. No jargon. No acronyms unless you define them.
MakeExecutive
Reformat for decision-makers. Lead with the recommendation, support with data, end with the ask.
MakeExecutive: Rewrite this as an executive summary. Lead with the recommendation. Support with 3 data points. End with the decision needed.
CompareOptions
When you have multiple paths, this lays them out side by side so you can actually choose.
CompareOptions: Give me 3 approaches to this problem. For each one: what it costs, how long it takes, what could go wrong, and who it is best for.
StressTest
Assume the plan fails. Figure out why before you commit to it.
StressTest: Assume this plan fails. What are the 3 most likely reasons? For each one, what early warning sign would I see?
Summarize
Cut everything down to what actually matters.
Summarize: Give me the 3 most important takeaways from this. Nothing else.
How to actually use this
Most people will read this and do nothing. Do not be most people.
Pick 3
Start with just three instructions:
- DefineOutcome
- ForceAction
- CheckGaps
These three cover most situations. Start here.
Stack them
Instead of asking:
"Give me marketing ideas"
You say:
DefineOutcome: Generate 50 qualified leads per week AddContext: B2B SaaS company, $15M revenue, selling to VPs ForceAction: Give exact steps to execute this week CheckGaps: What am I missing?
Same question, completely different output.
Iterate like an operator
Do not restart. Refine. Use these three instructions on any output to sharpen it:
- ImproveOutput
- Simplify
- MakeExecutive
Each pass gets sharper. Two or three passes usually gets you from draft to something you would actually send.
The shift
Most people ask AI for answers. That works sometimes. But the better mental model is directing a system to produce a specific outcome. You define what you want, give it the right context, tell it to act, then refine until the output is usable.
Do not overthink this. Pick one thing you are already doing with AI and add a couple of these instructions. See what happens.
Get the complete instruction set
All 15 instructions with examples in one place. Keep it handy.
Build your prompt library →Why this works: what the research says
Anthropic's prompt engineering research found that structured, context-rich instructions outperform ad-hoc prompts by 40-60% on task completion quality. The three biggest factors were specificity, relevant context, and output format clarity. Each instruction in this set targets at least one of those.
OpenAI's best practices documentation makes a similar point: breaking complex tasks into smaller steps produces better output than a single long prompt. That is the idea here. Each instruction handles one thing, and you stack them when you need more.
Frequently asked questions
Do I need to use all 15 instructions?
No. I would start with DefineOutcome, ForceAction, and CheckGaps. Those cover most situations. Add others when you run into a specific need, like comparing options or simplifying jargon. Most people end up using 5-7 regularly.
Does this work with any AI model?
Yes. I have tested these with Claude, ChatGPT, Gemini, and Perplexity. They work because they are about communicating clearly, not about any model-specific syntax.
What is the difference between this and prompt engineering?
Prompt engineering usually means crafting one careful prompt for a specific task. This is more like having a toolkit. Instead of writing something from scratch each time, you pick two or three instructions and combine them. Less creative writing, more assembly.
How is this different from the prompt library?
The prompt library has ready-made templates for specific tasks like sales calls or marketing campaigns. The instruction set gives you the building blocks to make your own prompts for anything. Think of the library as finished meals and the instruction set as ingredients.
Subscribe to the MrPrompts Newsletter
Join 5,000+ builders. One practical AI framework every week: prompt templates, workflow blueprints, and knowledge base strategies you can use the same day. Free.