Prompt Engineering, Project Briefs, and Practical Uses of ChatGPT

Two weeks ago, I completed Prompt Engineering Tutorial – Master ChatGPT and LLM Responses from freeCodeCamp with Ania Kubów. I wanted to do the course to learn more effective ways of writing queries to ChatGPT and other large language models, especially since I have previous experience using ChatGPT for work projects.1First-hand experience with LLMs is good, but why stumble through a new domain if there’s free information out there that can help you go farther, faster?

I found the course to be very useful, particularly because Kubów discussed the relationship between linguistics and writing effective prompts. Knowing the different aspects of linguistics is helpful because LLMs are basically high-octane statistics applied to linguistics.2Some folks with a deeper understanding of machine learning and large language models may disagree with this; however, I think this layman’s definition is sufficient for a working (but perhaps naive) understanding of how neural networks…well, work.

But the part that I found most helpful was the description of zero shot and shot prompting.

Zero Shot prompting is when you submit a query to an LLM without providing any additional context. For example, a zero shot prompt would be something like “Help me pack for a trip to Alaska.”

Few-Shot Prompting is when you provide additional context to the model for it to use in formulating a response. For example, a few shot prompt for a trip to Alaska would look like “I am planning a trip to Alaska with my dad in August. We will be visiting Anchorage, Denali National Park, and Fairbanks. Provide a sample itinerary for the trip focusing on must-see local destinations and provide a packing list for the projected weather.”3I took this trip IRL, and I did not use ChatGPT to help me pack. However, packing for Alaska can mean many different things based on time of year and planned activities. Are you going to Juneau or Nome? Will you be there for hunting, fishing, or something else? Ask generic questions; get generic answers.

Zero Shot and Few-Shot Prompting are both valid approaches depending on what your query is. However, my experience is that Few-Shot Prompting tends to provide a more detailed and useful response for more complex queries.

The day after I finished the course, I told my supervisor that I would write a project brief for our team to begin planning Easter services. If you’ve never written a project brief, it’s a staple document of project management that outlines the purpose, goals, milestones, and timeline for executing a project. It’s essential for stakeholder alignment, but it happens to be pretty boring to write.4Of course, there is an art to writing a good project brief or proposal, as it involves exercising strong economy of word in service of a very specific audience in order to highlight pertinent issues and questions related to the project. It’s like threading a needle that happens to have a lot of calendar and finance implications.

By the time I got around to writing the project brief, it was about two in the afternoon, which is about the worst time to start writing a key document for planning what is arguably the largest (or second largest) service that a church puts on all year.5The two biggest church services each year are Christmas and Easter. I decided that the best approach was to find a basic template that I could populate with my own information and data. But alas, I couldn’t find anything that quite fit my needs.6At the time of this writing, I’m in the process of building out document templates for things like project briefs, I don’t yet have an off-the-shelf solution for our organization.

Since I had just finished the prompt engineering course from freeCodeCamp, I was inspired to take ChatGPT out for a spin and have it write a first draft of the project brief. I wasn’t sure if I’d get what I needed, so I proceeded with the understanding that this was a test: It might work. Then again, it might not.

Here’s the prompt I used:

Write a project brief for a large Easter service taking place on March 31. Our services will be taking place on three campuses. Logistical details such as sound and audio and wayfinding signage need to be prepared and ready two weeks prior to the event. Social media and marketing assets need to be prepared and scheduled six weeks prior to the event. Sections in the brief should include Overview, Goals, Milestones, and Outstanding Questions.

I used the GPT 3.5 model, and I was very pleased with the results. Within a few seconds I had a working template that met my requirements. From just a single date, the model was able to calculate all of the project milestones for me and output them in the format that I specified. To be honest, putting together dates like that has been my absolute least favorite part of writing project proposals, briefs, and anything business that requires a list of dates. I enhanced the response with additional prompts to create a working schedule. For example, I had ChatGPT write the project milestones based on the date of the event like this:

Revise the provided brief with the following Milestones: Venue Selection 14 weeks prior to event, Audio/Video Company Contracted 10 weeks ahead of event. Branding design 10 weeks ahead of event, Marketing assets prepared 6 weeks ahead of event

After a few iterations, I had a working document that I could paste into Google Docs for team collaboration. The whole process of drafting the brief took maybe an hour or so.

I cleaned up the document by adding my own overview and inserting relevant details. Then I shared it with the team, and my colleagues immediately started adding relevant comments and data. Our finance team began adding budget details so we could make a decisions about potentially renting a satellite facility. Other team members added additional context and milestones. It was beautiful to watch the asynchronous collaboration happen without a single minute being wasted on a meeting.

By the time our scheduled Easter planning meeting came around a week later, we already had direction regarding the satellite facility, which itself was the critical decision for everything else related to the service, including communication, logistics, coordination with key partners, etc. Instead of starting the conversation at ground zero (i.e., “Well, what are we doing for Easter.”), we were able to come together based on a shared understanding of pending logistical decisions and outstanding questions. Most of the key questions that our team needed clarity on were answered within the first thirty minutes of a one hour meeting.

Now I can simply update the brief with the new decisions and plans, reshare it with the team, and begin working on executing the project plan. Project planning and knowledge management share one source in clear documentation, and this brief is an important part of strengthening both areas in our organization

The point is this: Using generative AI to create a project brief sped up collaboration and made for a much more effective meeting. It didn’t change the fact that we needed to have an in-person conversation—it just made the face-to-face discussion more fruitful.

Here are some key learnings I took away from this process:

  • Lots of off-the-shelf templates available online are simply marketing tools with highly refined SEO to get you to click on a specific page. Templates are useful only if they fit the intended context. Generative AI produced a template that fit my use case exactly.
  • Generative AI cannot do the real work of negotiation or collaboration, but it can help generate content and tools to support collaboration and discussion.
  • Garbage in, garbage out applies to code and to prompts for generative AI. The completeness and clarity of a prompt significantly influences the output of the model. I was very specific about the required sections, formatting for date milestones, and the ordering of dates, which resulted in a document that required minimal cleanup.

I’m learning that generative AI can be a powerful tool for speeding up workflows that hinge on team collaboration. I’m also learning the places where it makes sense to use it.

At the end of the day, I could have written that project brief from scratch. But would that have made things “better” in any way, shape or form? I don’t think so. What matters is using the tools available to enable and support human collaboration.

  • 1
    First-hand experience with LLMs is good, but why stumble through a new domain if there’s free information out there that can help you go farther, faster?
  • 2
    Some folks with a deeper understanding of machine learning and large language models may disagree with this; however, I think this layman’s definition is sufficient for a working (but perhaps naive) understanding of how neural networks…well, work.
  • 3
    I took this trip IRL, and I did not use ChatGPT to help me pack. However, packing for Alaska can mean many different things based on time of year and planned activities. Are you going to Juneau or Nome? Will you be there for hunting, fishing, or something else? Ask generic questions; get generic answers.
  • 4
    Of course, there is an art to writing a good project brief or proposal, as it involves exercising strong economy of word in service of a very specific audience in order to highlight pertinent issues and questions related to the project. It’s like threading a needle that happens to have a lot of calendar and finance implications.
  • 5
    The two biggest church services each year are Christmas and Easter.
  • 6
    At the time of this writing, I’m in the process of building out document templates for things like project briefs, I don’t yet have an off-the-shelf solution for our organization.
Scroll to Top