The effectiveness of AI is only as good as your prompts
There is no shortage of “use AI to drive efficiency” conversations, but did you know the secret sauce is in the prompts?
My aim in the next few months is to trial the use of different AI tools and how they can help make the work I do easier, and share my experience. And this is my first task.
I experimented with Chat AI and Chat GPT to find out how to write the perfect prompt.
My key takeaway:
While a deep and insightful conversation can happen with open ended questions (ie. What does happiness mean to you?), meaningful insights from AI require targeted and specific questions to get interesting insight.
AND, I still kinda suck at this. (but I’m practicing)
——
What I did:
To get the copy for the post I used ChatGPT and Copy AI to see what their responses were (I found ChatGPT had more interesting results so took the bulk of the content from it.)
But it was not a simple “ask once, get the content.” I had to keep prompting the tool for more detail, asking clarifying questions and diving into some of the detail of its response to get to some interesting nuggets.
I also found that I wanted to adapt the response to make it less “vanilla” and support any findings with my own Googling.
Here’s what I was able to discern from AI (and my experience with it), about the ingredients of a good and a bad prompt, using an example of wanting to do some research to define the opportunity space in the non-alcoholic market.
Examples of good prompts
"What are the most popular non-alcoholic beverage brands in Australia?"
Why it works:
This prompt is specific and clear, with a well-defined goal of identifying popular non-alcoholic beverage brands in Australia. It is also well-formatted and avoids any potential ambiguity.
"What are the latest trends in the non-alcoholic beverage market in Australia?"
Why it works:
This prompt is specific and clear, with a well-defined goal of identifying the latest trends in the non-alcoholic beverage market in Australia. It also provides context (the market in Australia) and is properly formatted.
"What are the main factors driving the growth of the non-alcoholic beverage market in Australia?"
Why it works:
This prompt is specific and clear, with a well-defined goal of identifying the main factors driving the growth of the non-alcoholic beverage market in Australia. It is also well-formatted and avoids any potential ambiguity.
Examples of bad prompts
"Tell me everything there is to know about non-alcoholic beverages in Australia."
Why it doesn't work:
This prompt is too broad and open-ended, with no clear goal or specific task for the AI model to accomplish. It is also too ambiguous and lacks proper formatting.
How it could have been better:
What are the new product launches or innovations in the non-alcoholic beverage market in Australia?
What are the growth opportunities in the non-alcoholic beverage market in Australia that brands should consider?
"What do Australians think about non-alcoholic beverages?"
Why it doesn't work:
This prompt is too broad and open-ended, with no clear goal or specific task for the AI model to accomplish. It also contains subjective language ("what do Australians think") that may not be easily interpretable by the model.
How it could have been better:
What are the common complaints or concerns that customers have about non-alcoholic beverages in Australia?
What features or benefits do Australian consumers value most in non-alcoholic beverages?
So what should your guiding principles be?
Keep it specific, better to ask a series of small questions than one big vague one
Don’t use jargon or slang, be precise and simple in your language
Try different tools and see the nuance in results - you still need to put the work in to create a meaningful piece of content to stitch it all together
You have to keep testing and learning, asking different prompts to get to the insights you need or to find the juicy detail
One prompt isn’t the silver bullet, but a series of good ones and you have some killer nuggets of insight
Be clear about the type of examples or data you need (ie. what stats do you need? how do you need them displayed?)
Want to get better at writing creative prompts?
Here are some of my favourite tools:
https://prompts.chat/
https://quickref.me/chatgpt
https://learnprompting.org/
Have you got any other tools/use cases should I explore next week around AI and how it can help research, ideation and growth tactic validation?