How I Use AI To Help Write Sermons
I write A LOT of sermons. Over the last 14 years, I've written an average of three full sermons every week. Often, in addition to that, I write a weekly Sunday School lesson. Each of these sermons includes hours of study and study notes, and a full manuscript which run between 2,700-4,000 words. If I was a "full-time" pastor, that would be a big lift every week.
But I'm not a "full-time" pastor. I also work in IT at a growing retail chain which just happens to be headquartered about 2 miles from my house and church, and I build software and websites to help churches like mine. I also have five kids, and I'm determined that they will not suffer for all that I have going on.
Over the last couple of years, AI has come to be a big help in writing sermons. But before I talk about how I use AI, I want to share a couple of principles that I believe are necessary guard rails for using AI in writing sermons.
Important Guard Rail Principles
1. AI cannot think for you.
The reason I write a full manuscript before I preach isn't so I can read my manuscript (I seldom do), it is so that I am forced to think through everything I am going to say. Writing is thinking. When you write, you clarify what you are thinking. If you cannot write it clearly, then you are not thinking clearly.
When we outsource our writing to AI, we are outsourcing our thinking to AI. While I may be ok with that in other areas, I'm not ok with that for sermons. People come to hear God's word, but they come to hear it through my genuine words and experiences - not the hallucinations of an LLM.
2. AI cannot study for you.
Building on the first guard rail, I don't think you can let AI do your studying for you. You have to read and think through the source material. You have to do word studies, you have to read commentaries, you have to wrestle with the meaning of the text. AI can aid in that - but there is a fine line between aid and delegation - and you cannot skip the hard stuff.
3. AI should not be plagiarized. (Or at least plagiarized very sparingly)
I'm against plagiarizing. When I use material from another preacher or commentary - I tend to give credit. I want people to assume that what I am saying is my words, not something ripped off of someone else. But AI was literally created to be plagiarized, it's job is to create content for you to use as your own.
A preacher can, with a simple prompt like this, have a full working sermon:
Pretend you are a conservative evangelical preacher trained in expository preaching under John MacArthur or Alistair Begg. Write a manuscript (around 3,000 words) on 1 Kings 10-11:11.
I preached on that text last nigh, and I just ran that prompt and what it returned was surprisingly solid. But even though it was created for my use, it isn't my sermon.
I guess the overarching principle here is that you shouldn't outsource the hard work of thinking and writing to AI. When you do, you are lying to your congregation, pretending that it is your intellectual labor, when in fact it is the work of an LLM.
But this doesn't mean that I don't use AI in my sermon writing. In fact, AI has helped me significantly and I use it every week.
How I Use AI in Sermon Writing
The way I think about the LLM is as a research and writing partner. It's as if I have someone in my study at all times, who is extremely knowledgeable, an extremely fast reader and an extremely fast typist who I can have an ongoing discussion about my sermon.
Whenever I write a sermon, I follow this basic flow:
- Decide on a text (usually sequential).
- Write a paraphrase of the text.
- Ensure that I correctly understand the text.
- Read lots of good commentaries and listen to sermons, taking notes as I do.
- Decide the main idea I want to preach.
- Develop the main idea into an outline.
- Decide on an angle for my sermon (a narrative hook).
- Write the manuscript.
- Preach the sermon.
I use AI as a partner for almost every step of that process. Typically, each sermon has two files:
- A notes files - where I do research and thinking.
- A sermon file - where I write my manuscript.
After I choose a text I setup a notes file like this:
# Summary and Outline
# Text and Paraphrase
# Themes
# Cross References
# Commentary Notes
Under each of those headings, I collect raw material for my message. After I have this notes file setup, I pass this prompt to the LLM (Currently, I use Claude but I've used ChatGPT, Gemini and Grok in the past and they all work well for this.)
I'm preaching this week on {Passage}.
- I like to start by making a paraphrase from the KJV and make sure that my understanding of the text is correct.
- Then I read about a half dozen commentaries and take notes.
- I decide what the thrust of my message is going to be and try to write it in one clear sentence.
- From all of that I try to write an outline.
- I decide on an angle to approach the manuscript and come up with an introduction.
- I write the manuscript, aiming to write 3,500 words or so (including the text.) I'm interested in feedback, but not feedback on pacing. I'll be sharing with you what I have, and using you as a writing and research partner through the process.
This sets up a conversation with the proper context and guardrails I need for that sermon. Then, as I go through each step, and as I write the manuscript, I pass what I have to AI to get help and feed back.
How AI Helps Me With My Paraphrase
I start my process with a paraphrase of the text. I read the KJV, and I always start by thinking through the text and putting it in my own words. As I finish a section, I will copy and paste it into AI. As I do, AI helps me in the following ways:
- It helps me see what I've mispelled.
- It alerts me when I miss a phrase or concept.
- It gently tells me when I don't understand something correctly.
It's actually that third case that has been most helpful for me. For instance, if in my initial paraphrase of 1 Kings 10:21, if I suggest in my paraphrase that the "house of the Forest of Lebanon" is Solomon's summer retreat in the woods. The LLM will kindly tell me I don't have that right, it's actually a nickname for Solomon's palace, given because of the huge number of columns in the house.
Through this back and forth, the weak spots in my understanding are exposed, and I then go back to study tools to make sure the LLM is correct and I am not (it almost always is). This probably happens at least once in every sermon I write - and it may be the most valuable thing the LLM does in the whole process.
How AI Helps Me As I Read and Study
After I do my paraphrase, I read and study commentaries (and sometimes listen to other preachers as I do other work.) As I do, I'm taking notes in my notes document of what stands out to me - summarizing my findings.
These notes aren't for the AI. They are for me. They aren't a book report, they are just the things that I found interesting or helpful.
After I have my notes, I'll often copy and paste all of them into the LLM and it does two things really well:
- First, it helps me identify themes across different writers.
- Second, it keeps those notes in its context for help with later steps.
How AI Helps Me Write A Sermon Statement, Outline and Preaching Angle
I think it is very important to have a single statement that drives the sermon. A one sentence summary of what you are trying to say. Call it a thesis. Call it a sermon statement. Getting that is usually my next step.
Of all of the parts of writing a sermon, this is the most mysterious. Most of the time, it just "comes to me." It is the product of all of the thinking and research I've already done.
As I write this out, I'll pass it to the LLM and ask for feedback. Often, the LLM helps me clarify parts of it or suggests ways to make it better, which I may or may not utilize.
From that sermon statement, I come up with an outline - a kind of plan of attack for the sermon. The LLM serves the same purpose, it's my writing partner, helping me think through what I want to say, and helping me clarify my ideas.
Finally, I'll come up with a preaching angle, which is my term for the rhetorical hook I plan to use as I approach the message. Again, I'll pass my ideas to the LLM for feedback, and use what I can.
How AI Helps Me Write My Manuscript
With all of this in hand (or in my notes document), I'm ready to write my sermon manuscript. (Which happens in another file.) But as I write it, I just constantly pass it back to the LLM for feedback and ideas.
As I get stuck, it might suggest a phrase or an illustration that is often very helpful. It will ask my what I want to do next and suggest things. The LLM knows my goals, it knows where I am going, and it knows all about the work I've put in so far - so it is the perfect writing partner. It basically eliminates "writers block" for me completely. It's like having a writing assistant in my study that never gets sick of brainstorming with me.
My biggest pet peeve with the LLM is that it is often sycophantic. Everything is "brilliant" or "perfect." LLM's are programmed to be encouraging - and sometimes they take it too far and that encouragement becomes meaningless. I could probably eliminate this with a tweak to my initial prompt, and will likely work on that in the future.
Honest Confession - it often writes my conclusions
As busy as I am and as much as I write sermons, the one place where I use the LLM in ways I don't like that go against my principles is when I'm running out of time and my conclusion isn't written. I have a bad habit of getting 90% done with an hour before I have to preach, and saying "I'm out of time, help me land this."
In the pulpit, I often don't read what it writes here. My instincts kick in and I say something off the top of my head. I don't feel good about it. But I guess its better than the alternative of having nothing.
Conclusion
AI has become an invaluable writing partner in my sermon preparation, but it works precisely because I've maintained clear boundaries about what it can and cannot do. The technology excels at being a research assistant, fact-checker, and brainstorming partner - roles that enhance rather than replace the pastoral work of studying, thinking, and writing.
For pastors considering AI in their sermon writing, I'd encourage you to start with the guard rails I've outlined. Use AI to enhance your study and refine your thinking, but don't let it become a substitute for the hard work of exegesis and theological reflection. Your congregation needs to hear from you, not from a language model.