Mr. Editor-in-chief How to Become an AI Power User with Better Prompts
Using AI well is no longer about typing a short question and hoping for a good answer. Modern AI models can research, reason, critique, write, analyze files, generate images, work with data, and even help build small applications.
The difference between a casual user and a power user is not magic. It comes down to how you give context, how you ask for reasoning, how you guide the model toward reliable sources, and how you iterate instead of expecting the first answer to be perfect.
This tutorial walks through practical techniques you can use immediately.
1. Stop Treating AI Like a Search Box
A beginner prompt often looks like this:
Write a good self-review for my boss.
The problem is that the AI does not know what you worked on, what your role is, what your company values, or what your boss expects. So the answer will probably be generic.
A better prompt gives context:
Help me write a self-review for my boss.
Context:
- My role: Automation engineer
- Major projects this year:
1. Improved testing station reliability
2. Helped debug Beckhoff TwinCAT integration issues
3. Supported customer acceptance testing
- Things I am proud of:
- Reduced downtime
- Documented troubleshooting steps
- Helped junior colleagues
- Tone: professional, confident, not arrogant
- Audience: my direct manager
Please draft a self-review based on this information.
AI works better when you treat it like a smart assistant who is capable but does not automatically know your situation.
2. Give the AI Enough Context
Context means all the information the model can use when generating its answer. This includes your prompt, previous chat history, uploaded files, pasted notes, screenshots, spreadsheets, and any tool results it can access.
A weak prompt:
Which apartment should I choose?
A strong prompt:
I am choosing between three apartments. Please compare them carefully.
Priorities:
- Quiet environment
- Short commute
- Safe neighborhood
- Reasonable rent
- Good internet
- Low risk of surprise fees
I will provide:
- Lease agreements
- Tenant reviews
- Neighborhood notes
- My budget
- Commute information
Read everything and think carefully before giving me a recommendation.
Please include:
1. Pros and cons of each apartment
2. Hidden risks
3. Questions I should ask before signing
4. A final recommendation
A useful rule: ask yourself, “What would a human expert need to know before giving me a good answer?” Then provide that information.
3. Ask AI to Think Hard on Difficult Tasks
For simple questions, AI can answer quickly. For complex questions, you should explicitly ask it to reason carefully.
For example:
I am comparing three cars. I will upload quotes, insurance estimates, maintenance notes, and feature lists.
Please read everything and think hard before answering.
Compare the cars based on:
- Total cost of ownership
- Insurance cost
- Reliability
- Safety
- Comfort
- Resale value
- Long-term practicality
Give me a clear recommendation and explain the trade-offs.
Modern reasoning models can spend more time working through complex tasks. Instead of saying “think step by step,” a more natural instruction is:
Think carefully before answering.
or:
Take your time and reason through the trade-offs.
This is especially helpful for decisions, planning, technical debugging, research, legal-document review, product comparison, and strategy.
4. Use Neutral Prompts to Get Honest Feedback
AI often tries to be agreeable. If you frame your idea as great, the model may unconsciously follow your lead.
Weak prompt:
I have a great business idea: mobile tie-dye service. Critique it.
Better prompt:
Please analyze the following business idea objectively:
Business idea:
Mobile tie-dye service for parties, schools, and community events.
Use this rubric:
1. Is there a real customer problem?
2. Is there enough market demand?
3. Is the business easy or hard to operate?
4. What are the startup costs?
5. What are the safety or logistics risks?
6. Does the idea have a competitive advantage?
7. What would make this fail?
Give it a score out of 100 and explain the score.
The second prompt gives the model a framework. It is less likely to flatter you and more likely to produce useful criticism.
5. Use Web Search for Fresh, Local, or Niche Information
AI models have built-in knowledge, but that knowledge may not include recent events or current details. Use web search when the answer depends on up-to-date information.
Use web search for:
- Current events
- Prices
- Product availability
- Laws and regulations
- Local businesses
- Opening hours
- Recent memes or trends
- Software versions
- Travel restrictions
- Highly niche topics
Example:
Please search the web and find highly rated gyms near Mountain View, California.
Requirements:
- Open now or likely currently operating
- Good recent reviews
- Reasonable membership options
- Within 15 minutes of downtown Mountain View
Please cite sources and mention when information may be outdated.
Do not rely only on built-in AI knowledge for things that change over time.
6. Guide the AI Toward Reliable Sources
Web search can return low-quality sources. AI may find Reddit posts, old blogs, marketing pages, or outdated pages unless you tell it what sources to prefer.
Weak prompt:
Are gray market peptides safe?
Better prompt:
Please research the safety of gray market peptides.
Use reliable sources first:
- Official health organizations
- Government regulators
- Peer-reviewed studies
- Medical institutions
Avoid relying mainly on:
- Seller websites
- Social media posts
- Forums
- Marketing pages
Summarize the risks, uncertainty, and what a cautious person should know.
For health, finance, legal, and safety-related topics, source quality matters a lot.
7. Use Deep Research for Complex Questions
Normal web search is good for quick answers from a few sources. Deep research is better when you need synthesis across many sources.
Use normal web search for:
What is the weather in Dubai this week?
Use deep research for:
Research how Dubai’s weather affects tourism patterns across different seasons.
Please consider:
- Temperature
- Outdoor activities
- Hotel pricing
- Major events
- Tourist comfort
- Seasonal demand
Use multiple reliable sources and produce a structured report.
Deep research is useful when the task would take a human many minutes or hours.
Good deep research tasks include:
- Market research
- Scientific literature summaries
- Travel planning
- Business analysis
- Policy comparison
- Product category research
- Competitive analysis
- Investment background research
- Technical tool comparison
Example deep research prompt:
Please run deep research on whether small manufacturers should adopt AI-assisted quality inspection.
Focus on:
1. Use cases in manufacturing
2. Cost and implementation difficulty
3. Required hardware and software
4. Benefits and limitations
5. Risks and failure modes
6. Case studies if available
7. Recommendations for a small factory
Use reliable sources and produce a structured report with citations.
8. Brainstorm Better by Giving Constraints
AI can generate many ideas, but basic prompts often produce generic ideas.
Weak prompt:
Help me build a workout plan.
Better prompt:
Help me design a workout plan.
Context:
- Age: 38
- Beginner level
- Equipment: 10 lb dumbbells, mini trampoline
- Available time: 15 minutes per day
- Problem: I lose motivation quickly
- Preference: fun, low-pressure routines
- Extra detail: I have a cat and want playful ideas
Give me 5 different workout plan options:
1. A normal beginner plan
2. A fun plan
3. A very easy plan for low-energy days
4. A plan using the trampoline
5. A habit-based plan using triggers
The more specific the context, the more useful the brainstorming.
9. Iterate Instead of Accepting the First Answer
Power users do not expect the first answer to be perfect. They use the first answer as a starting point.
Example workflow:
I need help paying down debt.
Context:
- Credit card debt: $1,100 at 19% interest
- Student loan: 8% interest
- Family loan: $900
- Monthly minimum credit card payment: $40
- I need to keep some cash available
Give me 3 to 5 possible repayment strategies.
Do not choose for me yet.
Then respond with feedback:
I do not like option 1 because it feels too passive.
I like the idea of paying down the 19% credit card debt first.
I forgot to mention that I have $450 cash coming soon, but I am also moving house next month.
Please create 3 updated plans based on this new information.
This back-and-forth helps the model learn what matters to you.
10. Use AI for Writing Without Producing “AI Slop”
A weak writing prompt:
Write a blog post about BlackBerry.
This often produces generic writing.
A better workflow is:
- Give notes and context.
- Ask for an outline.
- Critique the outline.
- Revise the outline.
- Expand into bullets.
- Critique again.
- Only then write the full article.
Prompt example:
I want to write an article about the rise and decline of BlackBerry.
Here are my notes:
[PASTE NOTES]
First, do not write the article.
Create a detailed outline.
Requirements:
- Make the argument clear
- Avoid generic business clichés
- Include historical context
- Explain why BlackBerry was strong
- Explain why it struggled later
- End with practical lessons for product teams
Then:
The outline is too generic.
Make the article more focused on product strategy and user behavior.
Remove the section about general smartphone history.
Add a section about keyboard loyalty and enterprise customers.
After the outline is strong:
Now expand this outline into a full article.
Use a professional but conversational tone.
Avoid sounding like generic AI writing.
Changing an outline is high leverage because a small outline change can improve an entire section.
11. Use AI to Edit Piece by Piece
Instead of asking AI to rewrite a whole article at once, work paragraph by paragraph.
Prompt:
Please critique this paragraph.
Focus on:
1. Clarity
2. Specificity
3. Flow
4. Whether it sounds generic
5. How to make it more useful
Do not rewrite yet. First explain what is weak.
Paragraph:
[PASTE PARAGRAPH]
Then:
Now rewrite the paragraph based on your critique.
Keep my meaning, but make it clearer and more natural.
This approach gives you more control and prevents the model from flattening your voice.
12. Use Rubrics for Objective Critique
If you want useful feedback, define the criteria.
Example:
Please critique this short story using the rubric below.
Rubric:
- Character development: 20 points
- Plot structure: 20 points
- World-building: 20 points
- Originality: 20 points
- Writing craft: 20 points
Instructions:
1. Score each category separately.
2. Explain each score.
3. Add the total score at the end.
4. Give 5 specific suggestions for improvement.
Story:
[PASTE STORY]
This is better than:
Please score my story out of 100.
A vague scoring prompt often produces flattering or poorly justified feedback. A rubric forces the model to evaluate more carefully.
13. Use AI Desktop Apps Carefully
Some AI tools can access files on your computer, read folders, rename files, edit documents, and organize projects.
A safe workflow is:
Look at the files in this folder and propose an organization plan.
Important:
- Do not move, rename, edit, or delete anything yet.
- First explain what you found.
- Then propose a folder structure.
- Wait for my approval before making changes.
After reviewing the plan:
The proposal looks good, but keep all original filenames unchanged.
Create folders only.
Move files into the new folders.
Do not delete anything.
Best practices:
- Run the AI tool only inside the relevant project folder.
- Do not give it access to your entire home directory unless necessary.
- Review permission requests carefully.
- Back up important files first.
- Be careful with delete operations because deleted files may not go to the recycle bin.
- Remember that AI edits may not preserve edit history.
14. Use Images as Input
Images can provide context that is hard to describe in words. You can upload screenshots, diagrams, handwritten notes, product photos, UI designs, charts, or error messages.
Example:
I will upload a screenshot of an error message.
Please:
1. Read the error carefully
2. Explain what it means
3. Identify the most likely cause
4. Give step-by-step troubleshooting instructions
5. Tell me what information you need if the first fix does not work
For design work:
I will upload a screenshot of a website design.
Please analyze:
- Layout
- Color palette
- Typography
- Spacing
- Visual hierarchy
- What looks professional
- What looks weak
- How to improve it
For handwritten notes:
Please read this handwritten note and turn it into a clean project plan.
Output:
1. Summary
2. Tasks
3. Deadlines
4. Open questions
5. Suggested next steps
15. Generate Images with Clear Visual Instructions
Image generation works best when you describe the subject, style, composition, and constraints.
Weak prompt:
Generate a robot.
Better prompt:
Create a clean cartoon-style illustration of a friendly humanoid robot sitting at a desk and typing on a laptop.
Style:
- Modern vector/cartoon illustration
- Warm office lighting
- Friendly and professional
- Suitable for a blog header
Scene:
- Robot wearing a navy hoodie
- Laptop on wooden desk
- Plants and bookshelf in background
- Window with daylight
- Focused working expression
Composition:
- 4:3 aspect ratio
- Robot centered
- Cozy workspace atmosphere
Image models can make mistakes with hands, text, and character consistency. For text-heavy graphics, it is often better to add text later using a design tool.
16. Use AI for Data Analysis
AI can help analyze spreadsheets by writing and running code behind the scenes.
Example prompt:
I will upload a spreadsheet of monthly sales.
Please analyze:
1. Total revenue
2. Revenue by product
3. Best-selling products
4. Month-over-month changes
5. Unusual spikes or drops
6. Possible business insights
Create charts where useful.
Explain your findings in plain English.
For personal fitness data:
I will upload my running data.
Please analyze:
- Weekly distance
- Average pace
- Longest runs
- Progress over time
- Signs of overtraining
- Practical recommendations
Use charts if helpful.
AI data analysis is not a replacement for an expert data scientist, but it is very useful for quick exploration, charts, summaries, and first-pass insights.
17. Use AI to Build Simple Apps
Modern AI tools can generate code for simple websites, games, dashboards, and utilities.
Example:
Build a simple web app that helps me practice basic French vocabulary.
Requirements:
- Show one English word at a time
- Let me type the French translation
- Tell me if I am correct
- Keep score
- Include a button for the next word
- Use simple HTML, CSS, and JavaScript
- Make the design clean and beginner-friendly
For a visual app:
Build a simple fireworks animation web app.
Requirements:
- HTML, CSS, and JavaScript only
- Dark night sky background
- Click anywhere to launch fireworks
- Random colors
- Smooth animation
- No external dependencies
AI-generated apps may need debugging, but they are excellent for prototypes and learning.
18. Know When to Start a New Chat
Chat history becomes part of the model’s context. This is useful when you are continuing the same task, but harmful when you switch topics.
Stay in the same chat when:
- You are refining the same article
- You are debugging the same codebase
- You are comparing options in one decision
- You want the model to remember previous constraints
Start a new chat when:
- The topic changes completely
- Old context may confuse the answer
- You want a clean, unbiased response
- You are moving from your own workout plan to someone else’s
- You are switching projects
A clean context often produces cleaner answers.
19. Troubleshooting Common AI Problems
Problem: The answer is too generic
Add more context.
This answer is too generic. Ask me 10 questions that would help you customize it.
Problem: The AI is too agreeable
Use neutral framing and a rubric.
Analyze this objectively. Do not assume it is a good idea. Use the rubric below and tell me the biggest weaknesses.
Problem: The AI uses weak sources
Specify source quality.
Use official, primary, or peer-reviewed sources where possible. Avoid relying mainly on forums, marketing pages, or social media.
Problem: The writing sounds like AI
Work from outline to draft.
Do not write the article yet. First create a sharp outline with a clear argument.
Problem: The model forgets important details
Restate the key constraints.
Before answering, summarize the constraints I gave you. Then use them in your recommendation.
Problem: The task is too large
Break it into stages.
First analyze the files and propose a plan. Do not execute yet.
20. Best Practices for Becoming an AI Power User
Use these habits consistently:
- Give the model enough context.
- Ask neutral questions when you want honest critique.
- Use rubrics for evaluation.
- Ask for multiple options instead of one answer.
- Iterate with feedback.
- Use web search for fresh or local information.
- Use deep research for complex synthesis.
- Start a new chat when old context becomes distracting.
- Ask the AI to think hard for difficult tasks.
- Review sources and outputs carefully.
- Use desktop file access cautiously.
- Treat AI as a thinking partner, not just a text generator.
Practical Prompt Templates
General Expert Prompt
Act as an expert assistant for this task.
Goal:
[Describe what you want]
Context:
[Provide relevant background]
Constraints:
[List limitations, preferences, deadlines, budget, tools, etc.]
Output format:
[Tell it exactly how to respond]
Before answering:
- Think carefully
- Identify missing information
- State assumptions
- Give a clear recommendation
Objective Critique Prompt
Please critique the following objectively.
Do not flatter me.
Do not assume the idea is good.
Use the rubric below.
Rubric:
1. Clarity
2. Feasibility
3. Risks
4. Market demand
5. Cost
6. Execution difficulty
7. Competitive advantage
Give:
- Score per category
- Total score
- Biggest weaknesses
- Specific improvements
Writing Workflow Prompt
I want to write an article about [topic].
Do not write the article yet.
First create:
1. A strong title
2. A clear thesis
3. A detailed outline
4. The intended reader
5. The tone
6. What to avoid
After I approve the outline, we will draft the article section by section.
Deep Research Prompt
Please run deep research on [topic].
Focus on:
1. Background
2. Current state
3. Main arguments
4. Evidence
5. Risks and limitations
6. Practical recommendations
Use reliable sources.
Compare different viewpoints.
Produce a structured report with citations.
Desktop App Safety Prompt
Analyze this folder and propose an action plan.
Do not edit, move, rename, or delete files yet.
First:
1. List what types of files you found
2. Propose a folder structure
3. Explain your reasoning
4. Ask for approval before taking action
Final Summary
AI becomes much more useful when you stop asking tiny isolated questions and start giving it the information a smart human assistant would need.
The core skills are simple:
Give context. Ask better questions. Use neutral framing. Request multiple options. Iterate. Use web search and deep research when needed. Let AI reason on hard tasks. Review its work carefully.
Used this way, AI becomes more than a chatbot. It becomes a research assistant, writing partner, brainstorming partner, analyst, design helper, coding assistant, and practical problem-solving tool.
Share this post
Related Posts
Software Fundamentals Matter More Than Ever in the Age of AI Coding
AI can generate code faster than ever, but strong software fundamentals are what keep that code understandable, testable, and maintainable.
Practical Codex Skills for Design Engineers
A practical guide to installing and using Codex skills, MCP tools, MagicPath, UI audits, and polishing workflows for better design engineering.
Build a Simple Obsidian “RAG-like” Knowledge Base with Claude Code
Build a lightweight knowledge base using Obsidian, Markdown, and Claude Code to organize raw notes into searchable, AI-friendly wiki pages without complex RAG infrastructure.
Using Firecrawl With Claude Code for Powerful Web Scraping
Learn how to use Firecrawl with Claude Code to scrape websites, bypass anti-bot protections, extract structured data, and build AI-driven research and automation workflows.