Generative AI Needs a Good Manager Too

What do great managers have in common with generative Al (GenAI) superusers? The ability to craft effective prompts with appropriate context and a clear request. The skills required to efficiently share context and articulate a clear ask (to a great degree) are transferrable from computer work to human interactions. And the core concept is as easy as following a three-step template.

Whether you are using GenAI to expedite a discrete task, or tapping into more complex functionalities, effective formulation of your requests is essential to generating high-quality outputs, and a poorly formatted prompt will result in a poor output. This is also true in your work with humans.

GenAI prompt templates provide you with various formulas for structuring information. Tapping into these templates can teach us how to more effectively leverage new Al tools and how to better communicate key information and expectations to our colleagues.

Prompt-ing

There are many GenAI platforms available and, increasingly, companies are even developing their own. For this article I am using ChatGPT to learn a bit more on effective prompting.

Me: What is your favorite prompt formulation for simple requests?

ChatGPT: For simple requests, I prefer prompts that are clear, concise and specific. Here's a basic structure I like:

• • Objective: Start with what you want to achieve.

  • Context: Provide any relevant background or details.

  • Action: Specify the action or output you're expecting.

This feels intuitive, which makes it all the better to serve as a new heuristic in your day-to-day work. Let's look at each of these steps more closely.

Starting With What You Want to Achieve:

You are far more likely to get the result you want if you clearly articulate the why. What need are you addressing? What is the purpose that should be top of mind?

In the case of Al this feels obvious — you often start with a clean slate; without stating the objective there is no known information regarding your goal. When sharing the objective, do your best to get to the heart of your purpose. Be as simple and clear as possible.

I asked ChatGPT for an example of an objective statement.

ChatGPT: "To maximize engagement and knowledge retention during virtual training for healthcare professionals."

This objective provides the purpose behind the action, helping align the request with the broader goal.

When working with people, alignment on the objective is often taken as a given. On the occasions where a deliverable comes back to you requiring significant adjustment, the disconnect may be fundamental; the requestee sees the goal differently from you. Take the additional step of checking for understanding, "... that is the high-level goal. Before I share more details, does that make sense?"

Alignment on the "why" will set up both Al and humans for success and create a strong foundation for you to build on as you introduce additional context.

Providing Relevant Background or Details

Your request did not fall out of a tree; it was generated from a unique set of circumstances. Understanding the background of a request is important to developing an appropriate response.

For a simple or generic request, you do not need to provide much context. For example, "create a numbered list summarizing the action items in these call notes." For a complex or highly specific ask, you will need to provide more. A non­ exhaustive list of examples includes:

• Provide a brief history of the events leading up to the request.
• Share any required inputs (such as raw data).
• Provide examples of related, completed work product (if applicable).
• Create a character (for instance, you are a learning professional with 25 years of expertise in training medical affairs professionals. You are passionate about supporting cross-functional collaboration and skill building.).

Take a moment to reflect on the information and experience you would pull from if you were to complete the request. What does your requestee (computer or human) know already that you should point them to? And where do you need to provide new information or guidance?

Where possible, leverage past information shared or experiences. Depending on the GenAI you are using, and the license that you are using it under, it may retain a long memory of previous requests. Or it may start fresh each time. For a person, you may need to re-send or jointly review information you have shared in the past to ensure it is top of mind.

I asked ChatGPT for an example. This is simple and high-level, but effective.

ChatGPT: "We're developing a virtual learning series for healthcare professionals who have limited time and diverse learning preferences. Past training sessions have struggled to maintain engagement."

This context provides the necessary background, helping me (or Al) better understand the scenario, audience and nuances of the request.

As you fill in gaps, be mindful that Al can process reference materials quickly so you do not need to be as judicious. You can share a lot of materials without significantly slowing down response time. Ensure that you remain aware of and adhere to applicable information sharing policies to avoid any confidentiality issues as some Al tools are confidential, and others are not.

In a high-profile incident, Samsung employees accidentally leaked sensitive information via the public version of ChatGPT. Today, many companies are setting up their own private GenAI where confidential information remains protected.

When working with people, it can be more difficult to strike a balance between providing enough context without overloading your requestee. Some people want more materials, some people need fewer, and in most cases, we want to be somewhere in the middle, with a digestible amount of information, yet no major gaps.

Sharing an appropriate amount of context with your requestee will help them to more effectively respond to the action you share.

Specifying the Action or Output Expected

Once the why and background have been established, share what the requestee needs to do. What specifically are you looking for as an output? Are there particular steps they should take to get there? Often, we open a request with action. In this formulation it comes last. Be as direct as possible about what you are looking for whether working with a person or a GenAI. Some ways to do this are:

• Specify your expected format
  • Are you looking for a bulleted list? An outline? A presentation? How many words?
  • Provide example work product.
  • Create a high-level outline or share a template where applicable.
  • If desired, ask for a few options and choose the output you like best.
• Detail the process, if appropriate
  • Note any calculations or processes that must be completed.
  • If there are areas of interest or focus, call them out!

I again asked ChatGPT for an example; this time focused on the action. It provided a very straightforward and clear ask.

ChatGPT: "Provide three strategies for making virtual training sessions more engaging and personalized for healthcare professionals."

This action statement clearly defines the expected output, making it easier to fulfill the request accurately.

Review and Refine

Once you have a great prompt — whether you are talking to a computer or a person — you will need to then review and refine the output you receive.

With Al, you can always ask it to try again (and again, and again... ). You will get the result very quickly, and you can be hyper direct, you won't hurt GenAl's feelings.

Recently I have used the below phrases to refine outputs. You can also rephrase or refine your original request and start over.

• "Use fewer words."
• "Try again using a bulleted list format."
• "Remove repetition."
• "Focus more on X."
• "You told me about A, I asked you about B. Try again."

As you work with people, the time to iterate is slower, and repeated revisions can be frustrating. Mind emotions as you offer feedback or request revisions. Try to be specific about what you liked and appreciated. Get curious and ask why they did something if it looks different from what you were expecting.

As you ask for revisions, link back to the objective of your request, e.g., "To better help our learners practice the tool and engage them more, might we add a fictional case study exercise here?" I could go into much greater detail about best practices for collaborating with people, but today we are thinking about the intersections of working with people and Al so I'll stop here.

Working With Humans vs. GenAI

As we close out, consider a few commonalities and differences when working with people versus GenAI (Figure 1). These apply from the request phase through completion. You'll note that with people, we are thinking about relationships, timeline and engagement as additional factors. With Al, we are thinking about the capabilities of the tool, and of compliance — please consult applicable IT policies on GenAI usage before you get started.

Strong Al prompting skills directly translate into better collaborative leadership and more effective interactions with people. Try out the objective, context, action formulation -    or adopt another template that works better for you (there are many to explore and experiment with).

Most importantly, practice and reflect on how you can continue to improve your prompting skills.

I'll leave you on a light note. I'm a big believer that learning should be fun, so why not practice my Al prompting skills with a low-stakes topic? I invite you to do the same.

Me: [Objective] I am hoping to jazz up some writing I am working on. [Context] The audience is members of LTEN and the focus is similarities between formulating great GenAI prompts and effectively collaborating with humans. [Action] Write a limerick summarizing the connection between these two topics.

ChatGPT: There once was a prompt, clear and bright,

That sparked an idea just right,

With humans or Al,

Good guidance will fly,

And teamwork turns insight to light.

Originally published in LTEN's Focus on Training magazine.