Should I Use AI to Help Me Write Sermons?

If a pastor decided he was going to outsource his sermon prep to ChatGPT or the like, here's what I'd expect:

1. Not having put in the time to interact with the text, I'd expect him to be reading a lot verbatim, stumbling over the words, and not making much eye contact.
2. Regarding the content of the sermon, I'd expect it to be fairly banal, a repetition of what others have said, and often in a way that could be wildly inappropriate--say Catholic interpretations being inserted into a sermon given for evangelicals.
3. Every once in a while, AI generates what is known as "hallucinations", so I'd expect that periodically, you'd have a whopper of heresy unloaded on the congregation.
4. When called upon to apply the text in counseling with a member, that pastor will tend to fall flat on his face because he's never learned to think on his feet.

Now perhaps we have too many pastors who already fall into these traps by simply parroting what the big church pastor says about the matter, but the reality is that the use of AI can make the situation far worse by isolating the "pastor" from the Word of God he's supposed to be expert on.

I’d say it depends a lot on what part of the prep process you use AI for.

I think it could be great for feeding it a text and getting a list of resources to look at, neatly organized. … which Logos has done for years without calling it “AI.” But now we call every algorithm that includes some machine learning “AI.”

It could also be useful for things like “Write me a good transition sentence between these two points.”

It could also be really good for guys who have a hard time with the concept of hierarchical outlines. A lot of people struggle with that, so they have a three point sermon, for example, and point 2 doesn’t have the same organizing principle as the the other two points. Or it half overlaps with point 1 or 3, etc.

This leads to messier delivery and saps energy and focus from the message.. and makes listening more work.

So ask the AI: “Can you see any ways to improve my outline?”

If it’s smart enough, that could be great for a lot of guys.

If the preacher is a manuscript preacher rather than extemporaneous, that opens a whole new can of worms for AI usefulness but also AI ethical problems. Probably also sermon quality problems.

One thing an AI can never do for a preacher—as Bert noted—is internalize the text for you. (This is like asking someone to eat for you.) In addition it can never grip you with why the text matters, and what propels you with a drive to communicate. Nobody who isn’t you—human or software—can do that for you.

So in a lot of ways it’s the same old problem preachers had if they tried to pull a sermon from a sermon collection book. If it’s not yours at the soul level, ,it’s just not going to work, even if it’s ‘objectively’ great.

The original question asked if it was ok to use AI as a help to prepare a sermon. As Piper expanded on it, he got to the place of asking if AI could write the whole sermon. I agree with Piper’s conclusion about that.

But as a tool for research, I think AI offers a lot of potential. It sounds like Piper is himself using it somewhat like that. It’s sort of like paying an assistant to do research, then publishing the resultant book and putting your name on it. (Anyone know of folks doing that?)

There are some interesting possibilities with it. GFA has a recent podcast on this (check back a couple of weeks on it). And see our current series on it at P&D. I think it’s worth thinking about.

Here are some questions I’ve thought of:

Would it be ok to submit an outline to an AI and ask it to improve the logic?

(Sometimes when I look back at an old message, I go, “What was I thinking?”)

What about using AI to find relevant illustrations? (We’ve got an upcoming episode touching on that.)

What about using AI to summarize a pile of articles on a subject to speed up your research? The AI might be able to give relevant citations. Sometimes I try to wade through a series of journal articles and find it a tough slog, especially if it is for just a single message. The GFA podcast touches on this.

...I am reluctant to endorse AI evem for the smaller parts of sermon prep that Don and Aaron mention. If a person needs AI to help with sermon outlines, research, and such, I've got to first ask "how did this person get through freshman English/rhetoric in college?", and second "is this person really qualified to be a teaching elder, apt to teach?".

It's really a rhetoric case of something I've observed ever since I was in college; when people start to rely on calculators to perform arithmetic, they lose the ability not only do to arithmetic, but also higher mathematical functions. Same thing will apply with AI-generated text.

A final note is that AI really doesn't do well at creativity--in effect "searching outside its current sample space"--and tends to be incredibly unreliable that way. To me, that's a deal killer for the pastorate.

And I bet a lot more preachers are using some aspect of AI in sermon prep than you realize. Logos has incorporated AI into its software. I haven't found much use for that as yet, but probably a majority of pastors use Logos.

And again, listen to the GFA podcast on this. Here's the link: Gospel Fellowship Association Missions (I was on my phone posting earlier.)

Yes, there is a creative process that must take place, but like all our tools, there is a right and proper way to use them.

BTW, think about how "non-AI" computing has improved writing and sermonic preparation. You could ditch the computer entirely, go back to a pile of paper books strewn all over the desk, use only pen and ink (or a quill) and do it the old fashioned way, the way creative juices "really" flow, or you could just use the tools the Lord has provided.

I use AI only for grammar (I use Grammarly). I honestly don't care if it is good at thinking. I want to do that.

I don't see AI wildly different than how pastors have used other resources like commentaries and study books. Some pastors even use the commentary to help build their structures. AI may not have the rigor that a researcher or author may provide to a work, so it needs to be used with caution, just like other sources like articles on the internet. If you are just solely using AI than you probably have another problem.

Don, I've watched how it works, and I actually work for one of the biggest names in AI in the world. I don't do AI for a living, but "hallucination" is actually a technical term for when AI gives a bad result. Yes, they trained me. Hallucinations are akin to when statistical studies give the wrong result because the person doing them was trying hypothesis after hypothesis to try and get statistical significance.

The simple reality is that there are a lot of things that happen when AI takes the place of actual study, and most of them are bad. I've seen it with calculators, and I've seen it with optimizers on engineering simulators, I've seen it with spellcheckers telling people to use the wrong word, and I've seen it a hilarious review of my personal blog from "Google Bard".

And anyone who does not already have a good comprehension of the subject at hand is going to be taken in when AI uses bad sources (a real problem) or hallucinates.

And anyone who does not already have a good comprehension of the subject at hand is going to be taken in when AI uses bad sources (a real problem) or hallucinates.

Obviously, but I don't think that is what anyone is advocating as a proper way to use the tool.

I encourage you to check the resources I mentioned earlier. There are ways to use AI legitimately that will save a lot of time and enhance the work one is already doing.

…I am reluctant to endorse AI evem for the smaller parts of sermon prep that Don and Aaron mention. If a person needs AI to help with sermon outlines, research, and such, I’ve got to first ask “how did this person get through freshman English/rhetoric in college?”, and second “is this person really qualified to be a teaching elder, apt to teach?”.

I don’t it see that way at all. Pastoral ministry is so multi-faceted, and the pulpit work is only about 1/3 of it (maybe 1/4: you have the people work, the worship work, and the planning and direction work).

As for research, it can be very time consuming or—with the right tools—less so, but the result is the same. You can do a 2 hour bus commute or do a 30 minute drive in your car, and you arrive in the same place.

As for outlines, I see it as a matter of levels of quality, and it’s not even close to the only quality factor. I’ve heard very effective speakers who’s outlines were a mess. They would be even better with a good outline, but in the list of qualifications for an elder, we don’t see “clear and logically flawless outlining skills.”

It’s just a matter of making “good enough” or even “more than good enough” and making “even better” in terms of helping listeners follow, understand, and be edified.

So, analogies…

• Research - walk, ride a bike, take a bus, or drive a sporty SUV? There are benefits to walking and biking sometimes. Not always.
• Outlining - you’ve got an old leg injury and it’s never going to heal; doesn’t mean you aren’t qualified to do neighborhood patrol, but you’re going to do it better with a cane or a crutch sometimes.

(I don’t know if there are even AI’s that do a decent job of either of these things yet, but it seems likely to happen before long.)

I’ll concede this though: Just like there are some benefits to walking or biking vs. taking your car, the work of research has some ancillary benefits. You learn things you weren’t trying to learn, and you find questions you weren’t asking. But for pastoral ministry… this is not always the right priority.

I've already commented on my use of AI in pastoral ministry on a previous thread. I use ChatGPT primarily to draft communications and to create study questions.

I do use Logos on a regular basis, and I have used their most recent AI search feature to conduct research on theological topics. This new AI search feature can look at every book in the Logos catalogue (not just your Logos library!) and summarize the results in a matter of seconds. That's crazy awesome. The old search feature in Logos was also incredibly powerful, but it was limited to your personal library, didn't really do a good job at selecting the most relevant sources, and didn't summarize the results for you.

Now, I can read the summary and see the sources that Logos AI used to build the summary. I can then select any of the sources used for the summary and read them separately for additional information. As with any resource, you must use discernment with the information presented.

So, I don't fear AI. I seek to use it to simplify and automate research tasks that would either be impossible for me (i.e. searching Logos's entire catalogue) or take a lot of time (i.e. searching every book in my library for a particular theological question or scriptural reference).

Logos or AI cannot substitute time spent in the Word and in prayer, which every faithful pastor should be doing. As D.A. Carson has observed,

There is no long-range effective teaching of the Bible that is not accompanied by long hours of ongoing study of the Bible. Effectiveness in teaching the Bible is purchased at the price of much study, some of it lonely, all of it tiring. If you are not a student of the Word, you are not called to be a teacher of the Word.

Inspired by how Don took a bible study and had AI turn it into a P&D article, I ran an experiment. I took a recent sermon transcript of mine and asked ChatGPT to turn it into a 1,500 word article with an easygoing, conversational but professional tone. The result is extraordinary. It is outstanding. I did some light editing and inserted some clarifications—but it’s very, very good.

ChatGPT didn’t make something new. It summarized a sermon I studied for and preached all on my own. I am wondering if there are ethical issues with posting the article, to be honest. I’m tempted to do it. I’m tempted to activate the “blog” page on the church website and turn the sermon into an article to post every single week, now. It’s that good.

Exactly. There are interesting ways to use this. It is often observed that fundamentalists don't write, and quite honestly, it is like pulling teeth to get some of our guys to contribute to P&D. I have several sources, and there always seems to be something, but one observation I have is that our guys are writing sermons every week. Often those sermons are on "cutting edge" topics as we pastorally guide and teach our people.

I think this is one way the tool of AI could be used to distill a verbal presentation into something useful for a wider audience.

For now, if I publish such articles, I'll add a disclaimer at the end, something like "summarized from a sermon [link] by AI" or some such.

I typically record my SS lessons on Zoom and then send out the link so people can watch if they happen to have missed the lesson. Zoom has been including some AI features recently, summarizing my lesson and listing Action Items.

Here are the action items from a lesson from Galatians 5:

1. All team members will actively work on mortifying the deeds of their body that they see in their lives, as listed in Galatians 5:19-21.

2. All team members will work on cultivating the fruit of the Spirit, as listed in Galatians 5:22-23, with a focus on love as the primary fruit.

3. All team members will pray for the Holy Spirit to work in their lives, helping them to love others and experience joy in every circumstance.

Tyler, did you pay for that?