AI Sermon Preparation: Drawing the Line Honestly

An AI-written sermon preached as your own is deception

Let us say the abuse case out loud before anything else, because most pages on this subject bury it. If you ask a chatbot for a sermon on Philippians 2, print what it gives you, and stand up on Sunday as though those thoughts came out of your study and your knees — that is deception. Not a gray area. Not a matter of liberty among brothers. Your people did not merely request forty minutes of true statements about the text; they gave their pulpit to a man, on the understanding that the man had been in the text and the text had been in him.

Notice that the wrong here is not mainly a copyright problem. No one's rights were infringed. That is why the plagiarism framing, though people reach for it, is slightly off the mark. The wrong is against the people who believed you did the work — and, quietly, against yourself, because a sermon is supposed to do something to the preacher first, and a sermon you did not prepare cannot.

Moreover it is required in stewards that one be found faithful. (1 Corinthians 4:2, NKJV)

Concede that fully and without qualification, and the rest of the conversation becomes possible. The question was never whether the abuse is real. It plainly is. The question is whether every use is that use.

If AI sermon preparation makes you uneasy, keep the unease

Talk with enough preachers and the same shape keeps showing up: quiet use and open discomfort, often in the same man. He has asked a chatbot to explain a participle, he has not mentioned it to anyone, and he is not certain whether he did anything wrong. We are not going to hand you a survey figure here — we do not have one we would stand behind, and a percentage would not settle your conscience anyway.

What matters is the posture. If you are uneasy, you are not behind the times and you are not a Luddite. The unease is doing exactly what pastoral instinct is for: it is asking who did the work. Keep it. The right posture toward a tool this capable is neither enthusiasm nor refusal — it is a settled rule you could state to another elder without embarrassment, and could still keep on a week when you are tired.

The line, task by task

Here is the part almost nobody publishes: which tasks fall on which side of the line, whatever tool you happen to be using. Not a principle to admire — a list to use.

Fairly delegated

• Locating cross-references and parallel passages you would have chased through a concordance anyway.
• Summarizing a commentary you own and intended to read, so you know whether to read it closely or skip it this week.
• Surfacing the structure of a passage — repeated words, inclusios, shifts in person or tense, the seams of an argument. Structure is observable. You still have to see it and judge what it means.
• Assembling background: what Colossae was, what a first-century household code assumed, what a denarius bought.
• Checking whether an illustration is worn out, or whether you already used it in March.
• Tidying grammar and cutting a manuscript that runs nine minutes long.
• Asking you the hard questions you would rather postpone.

Never delegated

• The interpretive judgment. What the author meant is yours to decide and yours to answer for.
• The doctrine, in your own words. If you cannot state the enduring truth of the passage plainly without help, you do not yet hold it plainly.
• The application to these people. No system has sat in the hospital room, buried the deacon's wife, or watched the young couple on the third row this month. It does not know your congregation. You do.
• The illustration from your own life. Borrowed feeling is the most obvious counterfeit in preaching, and your people can hear it.
• The conviction. A sermon that has not first done something to the preacher will not do much to anyone else.
• The words you say. Line by line, the manuscript in your hand should be yours.

One test covers most of the edge cases: would this sermon come out materially the same if a different faithful preacher ran the same tool on the same text? If yes, something has gone wrong, because the thing that was supposed to be irreplaceable — your reading, your people, your voice — has quietly dropped out.

How is AI sermon preparation different from using commentaries?

This is the fairest objection to everything above, and it deserves a real answer rather than a dismissal. You have leaned on Calvin, Henry, Stott, and Carson your whole ministry. Why is one kind of outside help legitimate and another suspect?

Three differences, and they are practical rather than sentimental.

1. A commentary has an author you can weigh. You know his tradition, his commitments, where he tends to overreach. You read him as a voice in a conversation and you push back. Generated prose has no author, no track record, and no accountability — yet it arrives in the same confident register as a scholar who has spent thirty years in the epistle.
2. A commentary argues; a generator delivers. Henry gives you reasoning you must accept or reject. A sermon generator gives you the finished form, already shaped, already outlined, already warm. The danger is not that it helps. The danger is that it hands you the destination before you have walked the road, and the walking was the point.
3. You already know how to credit a commentary. When you use another man's insight from the pulpit you say so — "as one writer puts it." The habit exists. Where AI is concerned the habit does not exist yet, which is precisely why the disclosure question below matters.

Why a general-purpose chatbot is Scripture-blind

There is a technical reason for caution that has nothing to do with theology, and you should know it before you trust any answer about a text.

A general chatbot predicts plausible language. Sermon-shaped prose about Scripture is easy to predict — the cadence, the three points, the phrase "the Greek here carries the sense of." Fidelity to the actual words sitting in front of you is much harder to predict. So when you ask about a less-traveled verse, you can get a paragraph that sounds like exposition but half-belongs to a neighboring passage: a clause quoted that is not in your text, a structure imported from a similar verse elsewhere, a lexical claim that would not survive two minutes with a lexicon.

It is not lying. It is guessing, fluently. Which means every claim about the text has to be checked against the text — and once you are checking every claim, you are doing exegesis, which is exactly where you belonged. This is the strongest practical argument that interpretation cannot be outsourced: not merely that it would be wrong, but that it does not actually work.

AI that asks; you answer

So the useful question is not how much a machine can say about a text — it is how little it needs to. That is what KERYGMA PRO is built on. It does not write your sermon and it does not hand you someone else's outline. It walks you through six stages and composes only from what you wrote at each one.

• The passage — you choose the text, and it stays in front of you the whole way through.
• Observation — what the text actually says: repeated words, contrasts, commands, promises, before you interpret anything.
• Interpretation — what it meant to its first hearers, and what the author intended by it.
• Doctrine — the enduring truth, stated in your own words.
• Application — what this asks of your congregation this week. You name it; the app does not know your people.
• Composition — a full draft in clear expository shape, built entirely out of stages one through five.

The questions come from the software. Every answer comes from you, and the finished sermon contains nothing you did not put there. Your work stays private to your own account.

It is in a Founding Beta at the moment — free to join, limited spots, three guided sermon creations included, so you can test the claim on a real text rather than take our word for it. It runs in a browser on Windows, Mac, Chromebook, or iPad, and there is a downloadable app for Windows as well.

Should you tell your congregation?

Use the plainest test there is: would your people feel deceived if they found out?

That question sorts most cases cleanly, because it separates a tool from a generator. Nobody discloses a concordance, a Bible software library, an interlinear, or a spell-checker. Those are instruments; the study is still yours. But if a machine supplied the interpretation, the structure, or the sentences, your people have been told something untrue by implication, and disclosure is owed — or better, do not use it that way.

If you want a line for a bulletin or a members' meeting, something like this is enough:

Sermon preparation here makes use of study software, including AI tools that ask questions, organize research, and check the manuscript. The study, the conclusions, and the words preached are the pastor's own.

And take note of the simplest diagnostic of all: if you find yourself hoping the question never comes up, you already know the answer, and no framework on this page will settle your conscience for you.

A rule you can keep

Let it ask. Let it fetch. Let it tidy. Never let it decide, and never let it speak.

That rule keeps the labor where Scripture puts it — on the man who will give account for the flock, who has been told to be diligent, to present himself approved to God, a worker who does not need to be ashamed, rightly dividing the word of truth (2 Timothy 2:15). Nothing about a new tool relocates that responsibility, and any tool that quietly tries to should be shut off.

If what you want is not a shortcut but a stronger method, the six movements above will carry you with nothing but a text, a pen, and the hours you actually have. Software can hold you to the order of them. It cannot do any one of them for you, and you should not want it to.