Prompt Engineering Tips: 3 Lessons to Get the Best from Generative AI Tools

Signal Boost: "I Feel Love", Donna Summer
A futuristic groove with precision and flow, I Feel Love captures the moment when great prompts turn into pure magic.

Talking to ChatGPT without context is like trying to order dinner by grunting at the waiter. You’ll get something off the menu, but probably not what you want.

I've worked with generative AI tools for a while now, and here’s the truth: the Large Language Model (LLM) is not psychic. It’s powerful, yes, but also polite. It waits for instructions, takes your vague ideas literally, and fills in the blanks with what it thinks you meant.

The good news? With the right prompting habits, you can shift from average results to genuinely useful, even brilliant, output. Here are the three biggest lessons I’ve learned.

1. Prime the pump: Ask it to ask you

Want great results? Don’t start by making demands, start with curiosity.

Begin every complex prompt with:
“Ask me questions one at a time to help you understand this request.”

This changes the dynamic completely. You’re no longer guessing what context the LLM needs. It’s drawing it out of you, like a thoughtful interviewer. Here's a quick example:

I needed help writing a product vision statement. My first attempt gave me generic fluff. Then I added: “Ask me questions one at a time to help you understand.” Suddenly I was in a back-and-forth conversation. The output? Spot on.

Why it works:

• You provide richer input
• The model narrows in on your actual intent
• You get a draft that’s closer to done on the first pass

2. Don’t stop at the first draft: Ask for improvements

Once you’ve got your output, don’t settle. Ask:
“How could this be improved?”

This simple follow-up unlocks a different mode in the model. One where it becomes its own editor.

I’ve seen it:

• Tighten up structure
• Suggest clearer language
• Replace clichés with original turns of phrase
• Match the tone more closely to my intent

For a leadership blog I wrote recently, the second version after asking for improvements was the one that actually captured my voice.

Bonus: you can loop this as many times as you like. Iteration is your friend.

3. Define the structure: Give it a format to follow

AI doesn’t just need what, it needs how.

Whenever I’m producing content like a blog post, I share the output structure upfront. For example:

• “Write in markdown”
• “Use a conversational tone that matches tone of voice”
• “Include an SEO title and a summary for metadata”
• “Add subheadings and a conclusion”

Even better, I’ve saved a reusable prompt template that does all of the above (you’ll find it at the end of this post).

The benefit? I spend less time fixing formatting and more time focusing on the ideas.

The Prompt Flow in Practice

Here’s the process I now follow when generating content:

Step 1: Ask the model to ask you questions one at a time
Step 2: Answer the questions and review the output
Step 3: Ask the model how it could improve the result
Step 4: Provide the desired output format and regenerate
Step 5: Repeat as needed as iteration beats perfection

What It All Comes Down To

Generative AI doesn’t require a PhD in promptology, it just needs good communication.

Treat the model like a curious collaborator. The more context you provide, the smarter it gets. Ask questions, welcome its feedback, and shape your prompt structure with intention.

That’s how you move from generic answers to gold-dust insight.

Reusable Prompt Template

I’m writing a blog post on the topic: “[INSERT TOPIC HERE]”.

Please use the following structure and format:

• SEO-optimised title at the top of the post.
• Summary paragraph to be used as a metadata description (max 2–3 sentences).
• Write the full blog post in Markdown format.
• Include clear subheadings for each section.
• Write in a tone that is [INSERT TONE e.g. conversational, insightful, lightly humorous, professional, etc.].
• End with a conclusion paragraph that wraps up the key message of the post.

Before you begin, ask me questions one at a time to gather the necessary context for the post. Once I’ve answered your questions and you’ve produced a first draft, I’d like you to review your own work and suggest how it could be improved, and then produce a tighter second version based on your suggestions.