AI Prompt Generation Tips for Churches

Depending on who you ask, AI is either in the process of revolutionizing life as we know it or fading quickly and quietly into obscurity. The truth, as with most things, is somewhere in the middle. There are definitely ways that today’s AI agents can be used to make people, even pastors, more productive, or give them a starting point. But they’re not foolproof, and they simply aren’t at a point (nor, I would argue, will they ever be) where they can replace the human element that’s essential to pastoral ministry.

Looking at Effective Use of AI in Ministry

Over the next several months, we’ll look at ways AI can be used effectively to aid pastors in their work. To be clear, this is not intended as a shortcut or to give rise to mediocre ministry where the robots do all the work. These are suggestions for how to use the tools available to augment, not replace, the essentially human work of caring for God’s people. As I often frame it at CTSFW, let the robots do what they’re good at so the humans can do what only they can do.

To that end, this month we’ll look at how to write a good prompt for today’s AI models. If you’ve tried today’s LLMs (ChatGPT, Grok, Gemini, etc.), you know by now they’re incredibly easy to use. But sometimes it can be quite difficult to get good information from them. Because these models do not rely directly on a fixed algorithm or process, the results may vary even when running the same prompt at different times. No prompt is perfect, and there’s no way to guarantee a useful result, but there are some things you can do to maximize your chances of getting the feedback you’re looking for.

Let’s work with the idea that we’re trying to come up with a game to play for tonight’s youth group meeting. I know we’re studying the parable of the Good Samaritan (Luke 10:25–37), but I have no idea what sort of game we might play to open up the night. I could ask, “Give me ideas for a youth group game,” and I’d get some okay results, but there are some simple best practices we can follow that will focus the model and ensure the results are what we’re after.

Define Roles

Start by defining the role you want the AI model to emulate. This helps focus the model on a particular area of expertise and narrows its vast content to a scope of responses based on that expertise. So, for example, if I were seeking to plan a youth Bible study around a particular text, I might begin the prompt with “You are an expert in youth ministry for current teenagers” before continuing with what I need the AI to do. By giving the model a set role, we eliminate much of the noise from additional content in its knowledge base. In fact, for this particular example, I might even narrow this further and focus it denominationally as well.

You are an expert in youth ministry in the Lutheran Church—Missouri Synod and are well-versed in current teenage culture.

Include Specific Details

The more contextual information the model has available, the better the result will generally be. By providing detailed information on the request, you can help the model tailor the results to your specific needs. For our example, assume we’re at a small congregation in rural Iowa. Let’s continue our prompt to include some of that contextual information.

You are an expert in youth ministry in the Lutheran Church-Missouri Synod and are well-versed in current teenage culture in the context of Shepherd of the Cornfield Lutheran Church in central Iowa. Your youth group consists of six youths, with an even mix of boys and girls. Most are from farming families, but two of the participants have parents who work for the local window manufacturing company in the engineering department. Tonight, the youth group will be studying the parable of the Good Samaritan from Luke 10:25–37 (ESV).

Request Clear Deliverables

But we need to let the model know what success looks like. To do this, we provide a very clear description of the thing(s) we want it to create. In general, the more specific you can be about what you’d like the model to produce, the better the end result will be. If you need an outline, ask specifically for an outline. If you want a more in-depth treatment, ask for a specific word count. Define the result, and the models will do their best to accommodate your requests.

You are an expert in youth ministry in the Lutheran Church-Missouri Synod and are well-versed in current teenage culture in the context of Shepherd of the Cornfield Lutheran Church in central Iowa. Your youth group consists of six youths, with an even mix of boys and girls. Most are from farming families, but two of the participants have parents who work for the local window manufacturing company in the engineering department. Tonight, the youth group will be studying the parable of the Good Samaritan from Luke 10:25–37 (ESV).

Prepare three suggestions for opening games you might play with the youth to begin the night. Games should relate to the Luke text without being artificially forced into context and should involve relatively little setup and no mess whatsoever. Provide a brief description of each game, a list of supplies, rules for play, and any tips or tricks for running the game effectively. The games should take no longer than 20 minutes from start to finish, including any explanation of rules.

Enforce Guardrails

In addition to letting the model know what to do, it’s often helpful to give it a few examples of ways you don’t want it to answer. Most models have built-in toxicity detection that prevents the most offensive content, but it’s still good to give clear guidance, especially in areas where it’s free to be creative or where it needs to stick to the plan. I added the following paragraph for this purpose. 

Do not depart from the scriptural text as found in the ESV. While the games do not have to be based on the texts, there should be a clear connection between the theme of the game and the theme of the text as understood by LCMS theology. All games must be appropriate for the youth group context and playable in a youth group made up equally of males and females.

Refine Results

The last phase of any engineering or design process is to test the product and see if it’s producing what we’re after or if we need to refine the design further to achieve it. It might be that what we have is good for now, and we want to use it as a template for a while and see how it goes. For now, we’ll run with the prompt as is. Let’s see what sort of difference expanding the prompt has made.

Here's the result from our initial prompt, "Give me ideas for a youth group game," (ChatGPT), put into a table for the sake of this blog:

ChatGPT then asked me, “Would you like me to suggest low-prep games (easy to run with minimal supplies) or bigger activity-style games that take more setup and time?”, to which I said yes with the refined prompt from above for this result. 

Here are three opening game ideas that connect naturally with the themes of the parable of the Good Samaritan (Luke 10:25–37, ESV). Each game highlights mercy, awareness of others, and acting to help—without being contrived.

Game	Neighbor Relay	Who’s My Neighbor?	Mercy Match
Theme Connection	The Samaritan acted quickly to help his neighbor, while others passed by. This game emphasizes noticing and acting.	The lawyer asked, “Who is my neighbor?” This game highlights recognizing people and paying attention.	The Samaritan showed mercy, not just words. This game emphasizes quick recognition of merciful actions.
Supplies	None	One blindfold	Prepared slips of paper with simple “helpful” or “ignoring” scenarios (e.g., “You see someone drop books—do you help?” vs. “You walk past someone crying in the lunchroom.”).
Rules	Divide youth into two teams of three. Arrange them in relay formation across the room. At the leader’s signal, the first person runs to the other side, taps the next person, then stays there. Here’s the twist: the person running must also “notice” a teammate who pretends to stumble (at random, a leader can call out someone’s name to “fall”). The runner must pause, help the “fallen” person up (quick hand-up), and then continue. First team with all members across wins.	Form a circle with one blindfolded youth in the center. Silently, the leader taps one youth in the circle. That person says, “Hello, neighbor” in a disguised voice (funny accent, whisper, etc.). The blindfolded player must guess who spoke. If correct, they switch places. If not, another round continues.	Scatter the slips face down in the middle of the group. Youth take turns grabbing one slip, reading it aloud, and deciding aloud whether it reflects showing mercy or ignoring a neighbor. If they get it right, they keep the slip; if not, it returns to the pile. Play until all slips are taken.
Tips	Rotate who gets called to “fall” so everyone plays both roles.	Encourage creativity in voices but keep it respectful.	Keep scenarios relevant to their lives (school, sports, farm chores, etc.).
Debrief	Who noticed? Who rushed? How easy was it to ignore someone?	Tie in afterward: It’s not always obvious who our neighbor is—but they are still there.	Which actions looked most like the Samaritan’s mercy?

 

Each of these games can be explained, played, and wrapped up in under 20 minutes. They require no setup beyond simple supplies, engage both boys and girls equally, and tie smoothly into your study of Luke 10:25–37.

While I’m not sure I’d use any of these games exactly as is, I could certainly take several of them and refine them quickly into a usable game for tonight’s meeting. Next time, we’ll look at taking these prompt generation principles and applying them to exegetical study of the biblical text!

Don’t miss Rev. Johnson‘s next blog on using AI for exegetical study! Subscribe to the CTS blog and get articles sent directly to your inbox.