Published on

Mastering AI Prompts: Frameworks

Authors
  • Name
    Callum van den Enden
    Twitter

Overview

This article is the first of many delving into best practices for prompting AI language models, focusing on structured approaches and leveraging known frameworks to optimise AI-human communication and task completion.

The Art of Prompting: Why It Matters

Let's face it, talking to AI can sometimes feel like trying to explain quantum physics to your cat. But it doesn't have to be that way. Proper prompting is the secret sauce that turns a frustrating AI interaction into a productive conversation.

Think about it. When you're chatting with a mate, you naturally provide context, explain what you're after, and tailor your language to suit them. It's no different with AI – except it's got the memory of a goldfish and the social skills of a hermit. That's where structured prompting comes in.

Structuring Your Prompts: The MECE Way

Now, I'm not usually one for fancy acronyms, but MECE (Mutually Exclusive, Collectively Exhaustive) is a gem when it comes to prompt structuring. It's about breaking down your prompt into distinct, non-overlapping parts that cover everything the AI needs to know. Here's how you might structure a prompt:

  1. Context: Set the scene. What's the background?
  2. Goals: What are you trying to achieve?
  3. Audience: Who's this for?
  4. Style & Tone: How should the AI 'speak'?
  5. Rules: Any specific dos and don'ts?
  6. Additional Information: Any extra bits that don't fit elsewhere?

This structure isn't just about being tidy – it's about giving the AI a clear roadmap. It's like giving someone directions: you wouldn't just shout out random street names and expect them to find their way, would you?

The Claude Quirk: Front-loading Additional Info

Here's a little insider tip for you Claude users out there. This AI has a bit of a quirk – it seems to pay more attention to information that comes earlier in the prompt. So, if you've got a load of additional info that's crucial for the task, stick it up front. It's like dealing with that one friend who zones out halfway through your story – get the important bits in early! Anthropic's documentation has a great prompting guide and even a course on prompt engineering. I highly recommend both.

Claude also loves XMLS, so if you're writing a blog post, you might want to include some XML tags to help the AI understand the structure of your content. For example, if you're writing a blog post about a new product, you might include the following XML tags:

<title>Introducing the New Product</title>
<author>John Doe</author>
<date>2023-01-01</date>
<body>
  <section>
    <title>Overview</title>
    <content>This is the overview of the new product.</content>
  </section>
  <section>
    <title>Features</title>
    <content>This is a list of the features of the new product.</content>
  </section>
  <section>
    <title>Pricing</title>
    <content>This is the pricing information for the new product.</content>
  </section>

This way, the AI can understand the structure of your content and provide more accurate and relevant responses. It means you can also reference specific sections easily in your prompt.

Leveraging Known Frameworks: Working Smarter, Not Harder

Now, here's where it gets really interesting. AIs like ChatGPT and Claude have been trained on a vast amount of information, including well-known frameworks and models. So why not use that to your advantage?

By framing your request within a framework the AI already knows, you're essentially giving it a head start. It's like asking a chef to cook you a specific dish rather than describing each ingredient and step. The AI can use its existing 'knowledge' as a springboard, focusing its processing power on the unique aspects of your request.

For example, if you're asking for a business analysis, you might structure your prompt around the SWOT framework. The AI already understands what SWOT means, so it can dive straight into applying it to your specific case.

Putting It All Together: A Practical Example

Let's say you want the AI to help you write a blog post about sustainable fashion. Here's how you might structure your prompt:

Context: You're a fashion blogger writing about sustainable fashion for your eco-lifestyle website.

Goals:

  1. Inform readers about the environmental impact of fast fashion
  2. Provide practical tips for building a sustainable wardrobe
  3. Highlight innovative sustainable fashion brands

Audience: Environmentally conscious millennials interested in fashion Style & Tone: Conversational and informative, with a touch of humour. Use British English.

Rules:

  • Avoid technical jargon
  • Include at least 3 actionable tips
  • Mention no more than 5 specific brands

Additional Information: -Use the '4 Rs of Fashion Sustainability' framework (Reduce, Reuse, Recycle, Reimagine)

  • Include a brief section on the concept of 'circular fashion'
  • Please structure the blog post outline using the '4 Rs' framework, with each 'R' as a main section. Under each section, include relevant information, tips, and brand examples.

See how this prompt gives the AI a clear structure to work with? It knows exactly what you want, who it's for, and how to approach the topic. Plus, by mentioning the '4 Rs' framework, you're giving the AI a familiar structure to hang the content on.

The Bottom Line

Effective prompting is all about clear communication. It's about giving the AI the tools it needs to understand and execute your request. By using a structured approach and leveraging known frameworks, you're not just making the AI's job easier – you're setting yourself up for better, more accurate results.

Remember, AI is a tool, not a mind reader. The clearer and more structured your input, the better the output. It's like programming, but instead of writing code, you're crafting prompts.

So, next time you're face-to-face with that blinking cursor in an AI chat, take a moment to structure your thoughts. Your future self (and the AI) will thank you for it.

What do you reckon? Have you tried structured prompting before? Got any tips or tricks of your own? Drop your thoughts in the comments below – I'd love to hear about your experiences wrestling with AI!