#50: How to Write Killer Prompts (Part 2)

Happy Monday everyone ???

It’s our 50th newsletter already, and we’re at 32,000 of you lovely subscribers. Pretty unbelievable stuff - thank you so much for all of your kind support.

So, are you enjoying the revamp and these new mini AI lessons??

Ready for another?

OK…let’s go.

Last week we learnt the foundations of writing a prompt:

• What bad prompts look like
• My 3 Cs framework for prompt writing
• A few simple examples

This week, we’re going to take it a step further, and dig deeper into the world of prompting, with loads more examples and some useful tips.

Proper Prompts are Pretty Big

The simple prompts we looked at last week were really short, so I could show you the 3 Cs in action.

As a reminder, those 3 Cs are:

Right, so now you remember the 3 Cs, let’s try applying it to a longer, stronger prompt.

This is a marketing prompt I use both for myself and for clients when I’m looking to build a tone of voice guide. It works so much better to analyse real examples and deduce tone of voice than just try and describe it.

Here, I’m using ChatGPT, but that can be swapped out for any LLM (Gemini, Copilot, Claude, etc.)

We can see the 3 Cs in action pretty clearly here. That first paragraph sets up the characters. Then we’re giving plenty of examples of content for context. And finally the whole prompt is very clear and specific, plus it’s broken down nicely into bullets and short sections.

We used a well-known framework in the above prompt (Four Dimensions of Tone of Voice, Nielsen Norman Group, 2016). Try throwing well-known theories and frameworks into your prompts - they work really well to add structure and flow to your results.

I use them all the time - and they inspire prompting use cases sometimes, too.

You’ll see we also used lots of different examples: “I'm going to feed you one piece of content at a time: an email, a LinkedIn post, an article, etc.”

Giving examples of what you mean is a crucial part of the Three Cs (Context), but if you don’t have absolutely perfect examples, feed in lots of different versions to develop a new amalgamated style.

Say, for example - you want to write a blog. But your company’s blogs so far are a bit…meh. Instead of including those, go through competitor sites and pick off blogs you think are great, and upload/paste those as examples instead.

Right, next. This one is a sales prompt, designed to help you refine your product/service offer.

Hopefully you’ll be able to see the 3 Cs in action a bit more now. This time the character that’s trained on the knowledge of a real person, Alex Hormozi (I know some people hate him, I think he’s great).

We’re also referencing a real book - this is working a bit like the theory/framework tactic from Advanced Tip 1 above.

LLMs have a huge problem. Around 15% of the time, they hallucinate (that means they make stuff up). That’s because they’re trained to predict the next word, not actually understand anything or apply logic. They’re also trained to be helpful and positive - they won’t ever say “erm…I don’t know”.

So you always need to edit responses. Always, always. Check every fact and assertion it gives you. You can make that much easier by asking for hyperlinked sources, like we did above. Remember, it can also hallucinate links, so sometimes they’ll lead to nothing!

Important: It’ll need the internet for this, so LLMs without internet access (e.g. Claude) can’t do this. ChatGPT, Gemini, Copilot…all good.

One of the best techniques to get strong results is to ask the LLM to ask you questions. It’s a great way to tackle prompts where you’re not exactly sure what data you should be giving it.

Say for example, you wanted to create a marketing plan for a new product, but you weren’t sure what information it needed to write a good one, you could add this section to your prompt:

And Finally…Weird Hacks

Did you notice an odd sentence in that last sales prompt? "Think carefully, take a deep breath and let’s begin”. Weird, isn’t it?

It’s actually been proven (here’s the rather techie link, if you’re interested) that adding certain statements gets you a more useful response:

• “Think carefully and take it step by step” - better answers
• “No yapping” - makes it more succinct

Something that also works is incentives and penalties. Try adding “I’ll tip you $200 if you do a good job” for more in-depth responses, or “if you include exclamation marks, I’ll penalise you $100” to successfully not do something. Try it - it actually works.

So that concludes your two-part mini lesson in writing killer prompts. Next week, we’ll be looking at the Google Killer and possibly the best AI tool in existence: Perplexity.

Until then, take care and have a fantastic week,

Heather

PS Want to work with me?

• Here’s my AI training brochure (prices going up in Sep)
• Check out my AI learning hub: just ￡25/mo
• Or just drop me a DM, I’m a friendly sort