I spend most of my life prompting AI and have come to a scientific conclusion.
There are basically three kinds of prompts I use:
- the “let’s see what happens” prompt
- the “do this exactly” prompt
- the “make it feel less like it was written by a microwave” prompt
Plus one bonus type: the prompt you use when you want the machine to surprise you.
That, for me, has turned out to be a more useful way of thinking about prompting than hunting around for the perfect magic phrase.
Because the important thing is not really the prompt itself.
It’s the mode you put the AI into.
Sometimes you want it to explore.
Sometimes you want it to execute.
Sometimes you want it to shape the feel of something.
And once I started thinking about prompting this way, everything got much simpler.
1. Exploration prompts
When you don’t actually know what you’re doing yet
This is my favourite mode.
Not because it produces perfect results, but because it often produces unexpected ones, which is generally where the good stuff lives.
Exploration prompts are deliberately loose. Instead of asking the AI for a polished answer, you invite it to wander around the idea with you.
For example:
I’m exploring video ideas around AI and WooCommerce.
Don’t give me polished titles yet. Help me explore interesting directions this could go. Where is the tension? What angles might surprise developers?
The important phrase here is “don’t give me polished titles yet.”
If you skip that step, the AI will immediately produce a list of extremely reasonable video titles that all sound like they were approved by a subcommittee for Strategic Content Excellence.
Exploration prompts work best when:
- you’re early in an idea
- you’re trying to find interesting angles
- you want the AI to challenge your framing
- you are still in the pleasantly chaotic stage where the plan is mostly vibes and optimism
Used well, the AI becomes something like a very fast brainstorming partner, except it does not need coffee and never says, “Let’s take this offline.”
2. Precision prompts
When you know exactly what you want
Once the idea is clear, exploration stops being helpful.
At this point, what you want is precision.
This is where I find AI particularly useful for things like YouTube scripts, because YouTube has a clear set of mechanics. It is not enough to have an idea. The video needs to pull people in, keep them there, and reward them for staying.
So instead of vaguely asking AI to “write a YouTube script,” I’ll prompt around the actual structure that tends to work:
- a strong hook
- an open loop early on
- a reason for the viewer to stay
- discoveries that escalate through the middle
- a clear payoff at the end
For example:
Write a YouTube video outline with:
• a strong 15-second hook
• an early open loop to create curiosity
• escalating discoveries in the middle
• a clear payoff at the end
If you just ask AI to “write a YouTube script,” it will often produce something with all the excitement of an instructional video about laminate flooring.
But if you give it a structure that already works, the results get much better.
At that point the AI is no longer guessing.
It is following a system.
That, for me, is one of the big lessons of prompting:
The best prompts often contain proven frameworks.
It is less like magic and more like handing a very fast intern your playbook.
3. Style prompts
When the content is correct but somehow still wrong
Sometimes the AI gives you something that is technically fine.
It is well structured.
It is grammatically sound.
And it has the emotional warmth of an airport self-check-in screen.
This is when you reach for style prompts.
Instead of changing the structure, you change the tone, energy, and feel.
For example:
Rewrite this introduction so it feels like a curious tech columnist noticing something interesting before everyone else.
Or for design:
Generate landing page ideas that feel like a modern fashion editorial. Clean layout, strong typography, lots of breathing room.
The content itself may not change very much.
But the feel changes dramatically.
And that matters, because people do not just respond to information. They respond to texture, rhythm, confidence, and taste.
Style prompts are essentially how you lend the AI your taste, which is important because taste is one of the few remaining human advantages, along with instinct and being able to tell when a paragraph sounds like it was assembled by a helpful microwave.
Bonus: the serendipity prompt
There is one more prompt I use quite often.
I call it the serendipity prompt.
This is the prompt you use when you want the AI to surprise you rather than simply obey you.
For example:
Based on everything we’ve discussed, what is the most interesting version of this idea that I might be missing?
Or:
Give me three directions for this article:
one obvious, one contrarian, and one that feels slightly strange but compelling.
These prompts are useful because sometimes the AI will suggest something that reframes the whole idea in a better way.
Not every time, obviously.
Sometimes it will also confidently suggest nonsense, which is part of the experience. Like having a very enthusiastic creative partner who has not slept properly in months.
But every now and then it comes back with something you would not have asked for directly, and that can be incredibly useful.
That is what I mean by serendipity.
Stepping back, the pattern looks something like this
| Stage | Prompt Type |
|---|---|
| Early thinking | Exploration |
| Building the thing | Precision |
| Making it feel right | Style |
| Looking for surprises | Serendipity |
The real trick
The biggest improvement in my prompting did not come from learning clever prompts.
It came from learning when to switch modes.
People often get stuck because they use the wrong type of prompt at the wrong time.
They ask for precision when they should be exploring.
They explore endlessly when they actually need to execute.
Or they forget to shape the style, and end up with something technically correct but oddly lifeless, like a beautifully organised spreadsheet about romance.
The trick is recognising which stage you’re in.
Once you start thinking this way, prompting becomes much simpler.
You stop trying to craft the perfect sentence.
Instead you just ask yourself one question:
What mode do I need right now?
The prompts behind this article
In case it’s useful, this piece was shaped with the same prompt modes it describes.
Exploration
Help me find a clearer framework for the kinds of prompts I’ve been using lately. I think there’s something around exploration, precision, style, and surprise, but I want a more logical division.
Precision
Turn this into an article called The 3 Types of Prompt That Are Working For Me Right Now. Use YouTube as the main example for precision, including hook, open loop, retention, and payoff.
Style
Rewrite this so it feels sharper, lighter, and more conversational while keeping the core argument intact.
Serendipity
What’s the most interesting ending for this article that I may not be seeing yet?
Which is either a satisfying way to close the loop, or a sign that I now spend far too much of my life talking to machines.
Leave a Reply