When it comes to AI models like Chat-GPT and its language-learning brethren, we’ve found one thing to be strikingly clear: you get what you give. Which means that prompt design – or what you say to Large Language Models (LLMs) and how you say it – is at the core of every AI interaction. That’s why we’re bringing you best practices and tips on how to speak AI from our most recent webinar, Talking To Machines.

Choose the right tool for the job

There are four different types of basic prompts:

  1. instructional (“write this”)
  2. question (“can you write this?”)
  3. input data or roleplay (“pretend you’re a writer and write this”)
  4. example (“write this so it sounds like x, y, and z”).

And different prompt types work best for different types of writing.

Ask it to roleplay as a specific person – like Barack Obama or Nicki Minaj if that’s your thing – when you need a strong tone of voice to catch attention in things like social posts. Or give it some examples of the kinds of tones of voice you’d like to sound like when you ask it to write a blog post draft. You can even include examples of how you don’t want to sound.

Better yet, you can combine prompts: give it examples of what you like, then ask it to roleplay. Who knew machines were so good at playing pretend?

Be precise with your words

If you’re writing something that has certain requirements such as word counts or need-to-haves like a call to action, be sure to include that in your prompt. And you’ll need to use firm, instructive language, such as “exactly”, as in “Write me an email that is exactly 200 words.”

Percentages are also your friends. If the LLM you’re using gives you something that’s tonally over the top (as they’re prone to do), ask it to remove 50% of the figurative language, or to cut the length by 25%.

Experiment, verify, then polish

Play with your prompts and see what you get. Because you’re never going to get something from it that’s perfect as-is – and that’s a good thing. AI is great at giving us jumping off points, rough drafts, even examples of how we don’t want to sound.

So experiment with whatever feels good, be sure to fact check anything as needed with a quick Google before publishing, and get ready to punch up, pare back, and polish whatever it spits out to give it that uniquely human touch.

Bonus tip: be kind

It’s worth it to throw in a please or thank you here and there while you’re prompting – it learns from us, so might as well lead by example.

Interested in learning more? Want to see some examples? Just want to brush up on your AI speak before the machine takeover? Check out the recording of our webinar or visit our dedicated page here.