This post picks up the trail and continues with more prompts and frameworks.

While I’d recommend going back to the previous post in this series to see where we started, it’s not required. In this post, we’ll continue exploring prompt rubrics or frameworks, giving examples on how to transform everyday questions into effective prompts.

Here we go!

CARE (Context-Action-Result-Example)

Humans learn best by personal experience. In AI-based scenarios, though, an AI platform isn’t going to be able to go out and have the experience of eating a sandwich or feeling the sun on its face. It can, however, learn by examples.

In the CARE prompt framework, you’re doing just that–providing the instructive context and guidelines and then providing an example of what a known good output looks like.

• Provide CONTEXT for the scenario
• Describe the task or ACTION the AI assistant should perform
• Clarify your expected RESULT or output
• Supply an EXAMPLE that shows the model what a suitable answer looks like

Without the prompt

In this example, we’ll ask the assistant to build a campaign for a pizza delivery company.

Not bad for 10 seconds of effort on my part. But what kind of results can we get with a better prompt?

With the CARE prompt

Suddenly, brand awareness.  And in the distance–the ringing of cash registers.

STAR (Situation-Task-Action-Result)

Many folks might be familiar with this acronym in the context of job interviews.  Frequently, in an interview scenario, a hiring manager is asking you to elaborate on a particular example as it relates to a question. For example, you might be asked, “Was there ever a time when you had to deal with a failed repair?” A suitable response would include a brief synopsis of the background (situation), the task you were expected to perform, the steps you took as part of the task, and what the outcome was.

This framework has a whole life outside of describing your experience to a potential employer.

The components of a STAR prompt are:

• Position a SITUATION or scenario
• Assign the model a TASK
• Provide additional or clarifying ACTION
• Define the qualities present in a the RESULT

Let’s see how it work!

Without the prompt

Suppose you are planning a family vacation. Without a prompt, your interaction with an AI assistant might go something like this:

As with other examples in this series, the model can come up with some great answers (even with little input detail).

With the STAR prompt

Of course, now it’s time to see how a more specific prompt can be used to get a more detailed response.

As you can see, providing a detailed list of results or expectations really pays off.

RISE (Role-Input-Steps-Expectation)

RISE is another framework that you can use and combines the role concept from other prompt constructs, the use of additional input detail or examples to set the context, and a request for a clear output with additional boundaries or guideposts in the form of expectations.

The RISE framework is comprised of the following elements:

• Frame the ROLE that the assistant should play
• Describe the background data to be used as INPUTS
• Ask the model for STEPS to complete a task
• Describe or define the goal or EXPECTATION of the output

Let’s say you’re working at an organization and you need to generate some ideas for a new subscription box service for cheese lovers.

Without the prompt

First, we’ll just ask a broad question about our nascent cheese subscription box without the framework, like we’ve done in the past:

And, like a lot of the examples before, we’ve gotten a decent high-level strategy, but not anything specifically actionable.

With the RISE prompt

Next, we’ll adapt the prompt using the RISE method by supplying specific details about our target audience and potential questions, ending by instructing the assistant to create a few pieces of content:

That’s a pretty stellar response (as someone who loves cheese, I am totally on-board with this cheese subscription box)!

Other ideas

Generative AI is a brave new world. So far, there aren’t any hard-and-fast rules about getting the exact responses you want each time. There’s a certain degree of randomness that comes into play. Just like asking 10 different people the same question can result in 12 answers, AI-based assistants are pretty similar.

I’ll recap some of my learnings over the last year or so here:

• There’s no definite right way to do this. It’s new.
• Feel free to mix and match prompt frameworks. You might get an idea using one framework that you really like–parlay it into a new prompt framework to help refine it!
• Another idea is to start with a broad, general prompt to start the creative process. Then, take some of the key thoughts in a conversational style and start drilling down into them.
• As you refine your ideas, provide more detailed guidance and specifics. You’ll get closer to the answers you’re looking for.

As the old saying goes, “unspecific data in, unspecific data out.”