The 3-Level Prompting System That Makes AI Insanely Useful

If you've ever felt like AI is just "not good enough," prompting is the fix.

It's a rare skill that's easy to learn and instantly useful— especially if your work involves teaching, writing, or any kind of thinking.

Until we reach general AI (we're not there yet), your results depend more on your prompts than on the model you use — even with Agentic AI.

That makes prompt design a highly valuable meta-skill today.

Whether you're using ChatGPT, DeepSeek, Gemini, or Claude (here's my breakdown on which tool to use when), your results are only as good as your instructions.

Over the past few months, I've gone deep into the prompt rabbit hole — taking expert-led courses, testing frameworks, and applying learning science to what works.

In this article, I'm sharing what I've learned structured into a 3-level guide.

You'll leave with replicable prompts that save you literally hours each week and a clear progression path from beginner to advanced prompt engineer.

✅ Level 1: The 5-Ingredient Prompt Framework

Here are the five core ingredients for Level 1 prompts that will turn AI into the sharpest thinking partner you've ever had. You want to use all five in every prompt.

1) T — Task

What should the AI do? Start with a persona, followed by a clear verb, and a specified output format.

Example: "Act as a cognitive scientist. Explain evidence-based techniques for long-term retention. Present the findings in a table, with columns for: technique name, description, evidence strength, and implementation effort. Sort the table from high impact, low effort to low impact, high effort."

2) C — Context

What details are needed? What's the end goal you want to achieve through AI's help? What's the impact you want to make?

The more context you provide, the better the output (this is especially true for reasoning models like DeepSeek and GPT's o1 o3 family).

Example: "You are writing for senior product engineers in their 40s who are learning how to learn — for their professional careers and their hobbies. Keep the tone clear, concise, and slightly witty. The goal is to make cognitive science approachable, so avoid jargon and use tangible examples."

3) R — References

Can you show an example?

When it's hard to describe what you want in words, show examples. Give the AI samples to mimic for tone, structure, and style.

Example: "Use a tone similar to this excerpt from Make it Stick: Practice that's spaced out, interleaved with other learning, and varied produces better mastery, longer retention, and more versatility. But these benefits come at a price: when practice is spaced, interleaved, and varied, it requires more effort. You feel the increased effort, but not the benefits the effort produces. Learning feels slower from this kind of practice, and you don't get the rapid improvements and affirmations you're accustomed to seeing from massed practice."

4) E — Evaluate

Is the result actually useful? Paste the received output into Perplexity or a fact-checking plugin.

Ask: Does this meet my goal? Is anything missing or incorrect?

For example, if the AI gave a generic article outline, ask it to compare it with real articles from [insert preferred blog or newsletter] — and point out the gaps

Once you've identified what is missing (you can also ask another AI LLM), go on to step 5: Iterate.

5) I — Iterate

Tweak and improve. Refine until the output meets (or exceeds!) your needs. Prompting equals iterating.

Here's a mnemomic I created so it's easier to remember all five steps in every prompt.

Practice Now

Open an LLM, take a learning goal or problem you're currently working on, and input the task, context, and references.

Extra tip: AI can accept and produce various types of content: Text, images, audio, video, and code. For example you can take an image of your fridge and pantry, upload it to ChatGPT (or any tool with image reading capability of your choice), and use the below prompt.

Example:

You are a world-class cooking teacher. Your task is to suggest three vegetarian recipes based on the ingredients in this image of my fridge and pantry. The recipes should be suitable for a family dinner, and take no more than 60 minutes to prepare.

Constraints: Vegetarian only; Prep time: max 60 minutes; Use only the ingredients shown (no shopping); Avoid overly complex techniques

_Output format: For each recipe, provide:

1. Name of the dish
2. Step-by-step checklist with estimated time per step
3. Ingredient list with quantities (grams/ml where possible)
4. Optional: Serving suggestion or tip to make it more special_

Be clear and concise in your tone, like a friendly professional chef guiding a home cook.

🔄 Level 2: Iterating with these 4 Techniques

Now that you've built a solid prompt, it's time to refine it. Great prompting isn't a one-and-done — it's iterative (remember step 5?).

Iterating is such an important step of prompting, and that's why we're dedicating level 2 to the ways in which you can iterate and improve your prompt.

In Level 2, we'll unpack how to simplify, reframe, reword, and constrain your way to better, sharper outputs.

1) Simplify

Break long-winded prompts into clean, digestible steps.

Why? Just like humans, AI performs better when it isn't overwhelmed. Complex logic or vague instructions often confuse the model.

❌ Instead of You're a UX designer. Create a wireframe for a homepage that clearly communicates our brand's values and unique selling points, while also prioritizing above-the-fold CTAs.

✅ Try this You are a UX designer. Describe a homepage wireframe that communicates brand values. Add a section focused on CTAs above the fold.

2) Shift the Perspective

Reframe the AI's role to unlock new types of thinking. Different personas yield different outputs — even for the same task.

❌ Instead of Act as a cognitive scientist and summarize this research paper in 5 bullet points.

✅ Try this You're a science journalist writing for Wired. Pull out the most surprising, counterintuitive insight from this study. Narrate them in a short paragraph with analogies and storytelling elements.

3) Modify the Language

Change the phrasing, tone, or structure of your request. If you're not getting useful results, don't just ask louder — ask differently.

❌ Instead of Make this sound better.

✅ Try this Rewrite this paragraph in the tone of Brené Brown — warm, vulnerable, but grounded in expertise.

4) Impose Constraints

Constraints create clarity. They force the model to get creative within boundaries — and often improve results dramatically.

❌ Instead of Suggest some book titles for my novel.

✅ Try this: Suggest 5 book titles under 5 words each using alliteration. Genre: speculative fiction. or Give 3 book titles that include the word 'mirror' and hint at psychological suspense.

Here's a mnemomic I created so it's easier to remember all four iteration steps in every prompt.

🚀 Level 3: Advanced Prompting Techniques

Most people just type a single question and hit send. That's called zero-shot prompting. It works fine for quick tasks, but if you want expert-level output, you need better techniques.

🔗 Prompt Chaining

Treat AI like a teammate.

You wouldn't ask a colleague to write a whole marketing campaign in one go — so don't do it with AI either.

Prompt chaining is the art of feeding one AI output into the next prompt — like building blocks.

Start simple. Then add layers. This helps you maintain control, reduces hallucination, and improve results through iteration.

Example sequence:

1. Summarize. Task: Act as a senior content strategist at a B2B SaaS company. Context: Summarize this article into 3 bullet points that highlight actionable insights for mid-level marketers in tech. Keep the language concrete and avoid buzzwords. Reference: Mirror the tone of Animalz or First Round Review — smart, practical, and slightly opinionated. Constraint: Each bullet must be under 20 words and include a key takeaway marketers can apply immediately.
2. Transform into hooks. Task: Turn each bullet point into a strong introduction hook suitable for a newsletter. Context: The audience is busy tech marketers. The tone should be punchy, confident, and curiosity-driven — think Morning Brew or Demand Curve. Constraint: Each hook must be tweet-length (under 280 characters) and make the reader want to keep reading. Use rhetorical questions or bold statements if useful.
3. Add Visuals. Task: Suggest 1 visual idea to match each hook that enhances reader understanding and retention. Context: These will be used in a Substack post or LinkedIn carousel, so aim for clarity and visual storytelling. Constraint: Each idea should be easy to execute (e.g., chart, diagram, emoji headline, or annotated screenshot), and must reinforce the takeaway — not just illustrate it generically.

🧠 Chain of Thought Prompting

Chain of Thought (CoT) prompting helps the AI reason through complex tasks by making it think step by step — just like solving a math problem.

This is especially helpful when: you want better answers, not just faster ones. The task involves trade-offs, logic, or multiple variables. Some advanced models now do this through enabling "reasoning" (like DeepSeek, Claude Opus, or Gemini), but it still helps to be explicit.

How to use it: Add this phrase to your prompt: 👉 "Explain your reasoning step by step before answering."

Example: Task: You're a learning designer with expertise in adult education and cognitive science.

Design a 4-week online course that helps busy professionals write clearly and confidently using evidence-based techniques. Your output should be an example outline with clear learning outcomes.

Context: The course should balance active practice, conceptual understanding, and peer feedback.

Constraints: Learners spend a max of 3 hours per week.

Reference: I've attached a PDF curriculum for an example format from a benchmarking course on learning how to learn.

Explain your reasoning step by step before answering.

🌳 Tree of Thought Prompting

Instead of getting one answer, ask the AI to explore several reasoning paths. This is useful when you're brainstorming or navigating abstract, creative, or strategic challenges. This lets you compare ideas side by side, avoid bias toward the first idea, expand your creative thinking with minimal effort.

Prompt formula: Propose 3 distinct options for [X]. Start by identifying the simplest possible version of this problem, then solve increasingly difficult versions. For each, explain pros and cons.

Example: Suggest 3 different structures for a 1-hour storytelling workshop. For each option, outline key activities, who it's best for, and one risk.

🪞 Meta Prompting

Ask the AI to write better prompts — for you.

When you're stuck, the best prompt… is a prompt about prompting. Meta prompting is perfect when you're not sure how to phrase your request, you're creating a prompt library, you want to level up your own prompting skills.

Example: Act as an expert prompt engineer. Write a prompt that generates 10 creative but practical startup ideas in the wellness space. The prompt should us the framework below. If something is unclear, ask me aditional questions. #Framework (copy and paste Level 1 and 2 of this newsletter edition).

Prompting = Thinking

Prompting trains you to clarify your thinking, ask sharper questions, and shape smarter outputs.

Join my community of Lifelong Learners to receive more helpful, science-based, tangible ways to learn with AI.

Resources

Davenport, T. H., & Mittal, N. (2024). What is agentic AI, and how will it change work? Harvard Business Review. HBR article.

Google. (2024). Google Prompting Essentials [Online course]. Coursera. https://www.coursera.org/learn/google-prompting-essentials

Hostinger. (2025). AI prompt engineering: How to write effective prompts. Hostinger Tutorials. https://www.hostinger.com/tutorials/ai-prompt-engineering

Learn Prompting. (2025). Introduction to prompting. LearnPrompting.org. https://learnprompting.org/docs/basics/introduction

Prompt Engineering Guide. LLM agents. Promptingguide.ai. Retrieved March 24, 2025, from https://www.promptingguide.ai/research/llm-agents

Sučik, Samuel & Skala, Daniel & Švec, Andrej & Hraška, Peter & Šuppa, Marek. (2023). Prompterator: Iterate Efficiently towards More Effective Prompts. 471–478. 10.18653/v1/2023.emnlp-demo.43