Art of Prompting with AI
A few months ago, I was helping a friend draft a funding proposal. She had ChatGPT open and typed:
“Write a proposal.”
The response? Generic. Bland. Useless.
Then I asked her to try this instead:
“Draft a one-page funding proposal for a youth literacy program in Detroit, focusing on community partnerships, measurable outcomes, and a $50,000 budget.”
The difference? Night and day.
If you’ve ever felt like “AI just doesn’t get it,” chances are the problem wasn’t the tool, it was the prompt. I’ve been working with LLMs (large language models) like GPT and Claude since their early releases, and here’s what I’ve learned:
Prompts are not just instructions. They’re conversations.
A good prompt is specific, grounded, and goal-oriented. A great prompt feels like setting the table before the meal, it prepares the AI to deliver what you’re hungry for.
In this lesson, I’ll walk you through the anatomy of good prompts, different types of prompting techniques, and the small adjustments that turn “meh” outputs into “wow” moments.
What makes a good prompt?
When I’m teaching new learners how to work with LLMs, I often start with this rule of thumb:
A good prompt is clear, specific, and instructs the AI on what, how, and why.
If your prompt is too vague, “Summarize this”, you’ll get a vague answer. If your prompt is overloaded, “Tell me everything about marketing, HR, training, and unicorn startups”, you’ll confuse the model.
Here’s a breakdown of what I call the Prompt Triangle:
- Goal: What do you want the AI to do?
- Input: What are you giving it to work with?
- Constraints: Length, tone, format, audience, etc.
Good Prompt Example:
“Summarize this 500-word article into three bullet points suitable for an executive who has no technical background.”
Poor Prompt Example:
“Can you summarize this?”
When your prompts include these elements, the LLM has the context it needs to help you, not guess at what you want.
Prompt structure and clarity
Let me give you a structure I’ve used for writing reliable prompts across multiple AI platforms:
Prompt Formula
Task + Context + Constraints + Tone (optional)
Example:
“Write a one-paragraph description (task) of a virtual wellness retreat (context) in under 75 words (constraint), using friendly and uplifting language (tone).”
This structure isn’t mandatory, but it’s wildly helpful, especially when the default outputs feel robotic or off the mark.
A few other quick examples from my own use:
- “Convert these notes into a LinkedIn post with a confident and casual tone.”
- “Translate this customer support reply into Spanish, keeping the tone apologetic but clear.”
- “Rewrite this academic paragraph in plain English for a non-technical audience.”
Once you practice this structure a few times, it becomes second nature.
Types of prompts
Now let’s talk strategy. Not all prompts are created equal. There are three key styles you should know about:
1. Zero-shot prompting
You give the model a task without any examples.
“Summarize this article in one paragraph.”
Useful for: simple or well-understood tasks.
Limitations: LLM may interpret ambiguously.
2. Few-shot prompting
You provide examples to teach the model how to respond.
“Convert the following formal messages into casual tone.
Input: ‘Dear team, I hope this message finds you well…’
Output: ‘Hey everyone, just wanted to check in…’
Now, convert this message…”
Useful for: formatting, tone shifting, specialized outputs.
3. Role-based prompting
You assign the AI a persona or role.
“You are a career coach. Write a motivational email to someone struggling with job applications.”
This style is shockingly effective. I’ve used it to get feedback on lesson plans, simulate peer reviews, and even test UX copy with pretend customers.
Try combining types:
“You are a resume expert. Based on the example below, rewrite this job description to sound more results-oriented.”
📣 Pro Tip:
When in doubt, tell the model what it is. It frames the conversation.
Summarizing vs. creating
A common mistake I see: using the same prompt style for totally different tasks.
Let me explain.
Summarizing tasks work best with:
- Word or bullet limits
- Defined audience (“Explain this to a CEO…”)
- Tone control (e.g., casual, formal, executive)
Prompt:
“Summarize this Slack conversation into a 3-bullet action plan.”
Creative tasks need:
- Context about the purpose
- Freedom to explore (but clear direction)
- Optional: formatting structure (e.g., “as a poem” or “as a social caption”)
Prompt:
“Brainstorm 5 playful Instagram captions for a plant delivery startup.”
Treat summarizing like a reduction. Treat creative work like a collaboration.
Iterating and refining outputs
Even seasoned prompt writers don’t get perfect results on the first try. That’s why feedback loops matter.
Here’s an actual example from a DevsCall lesson draft:
Initial Prompt:
“Write an introduction for a tutorial about Python for beginners.”
Result:
“Python is a popular programming language used by many…”
Yawn.
So I replied:
“Make it more conversational and add a relatable hook for someone switching careers into tech.”
Improved Result:
“So you’re thinking about switching careers, and someone told you to learn Python. Great news: they’re right. Let’s break it down together…”
Now we’re getting somewhere.
Use follow-up prompts like:
- “Make this more concise.”
- “Add examples.”
- “Change the tone to friendly but professional.”
- “Turn this into a listicle format.”
- “Explain it like I’m in high school.”
The beauty of LLMs is their willingness to revise tirelessly. Don’t be afraid to ask for more.
Prompt pitfalls to avoid
Over the years, I’ve reviewed thousands of student prompts, and I’ve seen some common traps. Here are a few to steer clear of:
1. Vague instructions
“Explain this.” → Explain what? To whom? For what purpose?
2. Overstuffed input
If you paste 6,000 words and say “summarize,” the model may freeze or hallucinate.
Break input into smaller chunks. Be specific about what you want from each section.
3. Leading prompts
Avoid stacking your assumptions into the question.
“Why is Python clearly better than Java?”
“Compare Python and Java for web development use cases.”
4. Mismatched tone
You ask for a friendly tone and then paste a legal disclaimer as the input. AI will struggle to find the tone it’s not shown.
Match your prompt tone to your desired result.
Write and refine 3 prompts
Let’s practice what we’ve covered.
Try this:
- Write a zero-shot prompt
E.g., “Summarize this blog post in three bullet points.”
- Write a role-based prompt
E.g., “You are a copywriting expert. Rewrite this homepage headline to sound more confident and modern.”
- Write a creative task prompt
E.g., “Brainstorm 10 name ideas for a mobile app that helps people organize their bookshelves.”
Now… refine each one.
- Can you clarify the audience?
- Add a tone requirement?
- Provide an example?
- Cut ambiguity?
This is how you build prompt muscle. It’s like learning to write better emails, you get better by doing.
Conclusion
If you remember one thing from this lesson, let it be this:
The prompt is the steering wheel. The model is the engine.
When you learn to steer well, AI tools become extensions of your creative process, not just tools you use, but partners in productivity.
So don’t just type. Think. Test. Tweak. And treat your prompt like you’d treat a conversation with a colleague, it needs context, direction, and clarity.
Your next assignment won’t be to master AI. It’ll be to speak its language, and that language starts here.