One of my students bounces up to me, practically vibrating with excitement.

I have that effect sometimes. Ha! No, they had a new business idea and they were buzzing.

"Kyle, I've got it! I'm going to build an AI tool that automatically scrapes social feeds, converts articles into every format imaginable, generates videos, and handles all my social media. It'll be amazing!"

Yeeeeeaaahh….sorry, no.

Not because it's a bad idea - it's actually brilliant. But it reminded me of my early days, trying to boil the ocean with my first AI projects.

Charging in trying to DO ALL THE THINGS and then being disappointed.

Here's what I've learned: the path to that amazing end-goal starts with nailing one simple transformation first.

Let’s get that done before moving up the levels of AI.

Focus First, Scale Later

That student's social media automation tool? Let's break down why it's both brilliant and problematic as a first project:

Brilliant because

• Clear value proposition (save time on social media yay!)

• Solves a real problem (content repurposing is hella time-consuming)

• Potential for monetisation (read: people and businesses would pay for this. We want that)

Problematic because it's actually multiple tools in one

• Content scraping system

• Text transformation engine (article → various formats)

• Video generation tool

• Social media scheduling system

• Engagement automation

Each of these could be a product in its own right. Trying to build all at once is a recipe for frustration. Instead, let's pick ONE transformation and nail it. In this case, maybe start with "news article → video script". One input, one output.

It’s still valuable. It’s just more focused.

Instead of doing a whole bunch of stuff poorly we’ll do one thing well.

Writing Your First Prompt

This is where most people rush in with something like: "Turn this article into a video script"

And then wonder why the results are... meh.

That’s a PEBKAC issue : “Problem Exists Between Keyboard And Chair”.

Let's build this properly. I've developed what I call the RISEN™ approach. Let me show you how it works with our article-to-script example:

Role: Give your AI a clear identity

"You are an expert video script writer specialising in transforming news articles into engaging social media video scripts."

Instructions: Set clear guidelines

"Your task is to transform news articles into 60-second video scripts that hook viewers in the first 3 seconds. Focus on one key message."

Steps: Break down the process

"1. Read the article and identify the most compelling angle 2. Create a hook that grabs attention in 3 seconds 3. Structure the main points in visual scenes 4. Add transitions between key points 5. End with a strong call to action"

End Goals: Define success

"The final script should

• Be 60 seconds when read at normal pace

• Start with a scroll-stopping hook

• Present one clear message

• Use conversational language

• Include visual scene descriptions

• End with viewer engagement prompt"

Narrowing: Test and refine

 This is where we add constraints based on testing. Ie. we run the prompt, check the results and then go back and add fixes. Maybe we find our scripts are too long, or the hooks aren't grabbing attention. We adjust: "Hook must be a question or surprising statistic" or "Don’t include stage directions, only the script of the text”.

Testing Your Prompt

Before you even think about building anything fancy, get this basic prompt working!

I can't stress this enough: If you can't get good results at this level, adding more complexity won't help.

Adding more crap to weak foundations will not help.

A fantastic tool for testing prompts is Anthropic's Console (console.anthropic.com). It gives you a clean interface for:

• Testing prompts systematically

• Tracking what works and what doesn't

• Refining your approach based on results

It’s a more advanced tool but well worth getting used to and adding to your arsenal - especially when building prompts for applications (that will be used again and again and again!)

What If It's Not Working?

Sometimes you'll find that even with a well-structured prompt, you're not getting the results you want. This usually means one of two things:

1. Your transformation is too complex. This is the most common problem that I see when I sit with students. Too much! Solution: Break it down further. Instead of article → video script, maybe start with article → key points → script structure → final script.

2. You're asking for something beyond current AI capabilities. Less common but sometimes AI just isn’t up to the task! (Yet) Solution: Adjust your expectations or change your approach. AI is not necessarily the best tool!

Moving to Level 2: Projects

Once your basic prompt is working reliably, we can level up to a proper Project. This is where it gets interesting - because we can start adding context that makes our AI tool smarter.

In Claude and ChatGPT, we can create a “Project” that includes

• Your core prompt

• Knowledge base

Think of a knowledge base as your AI's reference library. Just like you might have a collection of resources you refer to when working, your AI tool needs its own set of references to produce better results.

Obviously AI can technically access most of the world’s information (AIs were trained on the internet and generally have access to the internet too).

BUT sometimes we want to specify what knowledge we want our AI tool to draw on. This is where we use a knowledge base - to tell the AI “OK, I understand you know nearly everything. But for now here’s what I want you to focus on”

For example for our news article to short video tool it might include

• Your core prompt

• Knowledge base of

  • great video scripts

  • hook examples that actually work

  • best practices for script writing

  • etc.

But here's the key - don't just dump everything you can find into your knowledge base. The more you give it the less importance it’ll ascribe to each individual item. Instead, be selective. Include:

1. Examples. Find the best examples of what you're trying to create. For our script tool, I'd include scripts that actually went viral, not just any old scripts.

2. Pattern Templates. Look for recurring patterns in successful outputs. In video scripts, maybe certain types of hooks consistently perform better. Add these patterns as templates.

3. Technical Guidelines. Include any technical constraints or best practices. For video scripts, this might be optimal word counts for different video lengths, or rules about scene pacing.

4. Common Pitfalls. Also document what doesn't work! If certain approaches consistently fail, include these as anti-patterns to avoid. This is super powerful especially as content you add over time as the tool is used more.

If still in doubt use this

You are an AI specialised in analysing prompts and suggesting knowledge base enhancements. Review my current prompt:

[Paste your prompt here]

Identify areas where additional context or examples could improve results. Consider:
- What reference materials would help consistency?
- What examples would improve quality?
- What templates could speed up the process?
- What best practices should be included?

Remember, you don't have to move to Level 2 (Project) if Level 1 (Basic Prompt) is working for you!

Some of my most-used tools are still just well-crafted prompts in ChatGPT that I have saved for easy use. That’s perfectly fine! But if you want more consistency and power, Projects are your friend.

In Part 3, we'll look at taking your working prompt and turning it into a custom AI assistant using tools like Launch Lemonade. But for now, focus on getting that first prompt right. Without this foundation, nothing else matters!

Keep Prompting,

Kyle