AI Sermon Generators vs. AI Sermon Research: A Pastor's Guide

Start with the numbers, because they will probably make you feel less alone. Barna Group's 2026 research with Gloo found that 87% of pastors now use AI in some form. Half use it for brainstorming, 36% for biblical and theological research, 34% for small-group discussion questions — and 24% use it for sermon writing or editing, double the 12% of just two years earlier. At the same time, 71% of pastors describe themselves as cautious about it.

In other words: if you are using AI in ministry and feeling conflicted about it, you are not a laggard or a compromiser. You are the statistical majority of working pastors. What most pastors are missing is not permission — it's a clear line between the AI work that strengthens preaching and the AI work that hollows it out. That line is what this guide is for, and the short version is a distinction: AI sermon research is a different activity from AI sermon generation, and almost everything hangs on which one you're doing.

An AI sermon generator does what the name says: passage and topic in, outline or full manuscript out. The output is typically competent — clean structure, plausible illustrations, orthodox-sounding points. The technology works. The problem is not quality. The problem is what a sermon is.

A congregation listening to you preach is making an assumption so basic nobody states it: that the words are yours — that the text passed through your study, your prayer, and your convictions before it reached their ears. A generated manuscript delivered without disclosure quietly breaks that assumption. It isn't plagiarism in the legal sense; the machine doesn't mind. It's something closer to a broken promise about who is actually speaking.

The common defense deserves an honest hearing: 'I don't cite my commentaries from the pulpit either — how is AI different?' Here's how. A commentary is a known, authored, reviewed work; when you draw on it, you are weighing a named scholar's argument and taking responsibility for adopting it. Generator output has no author, no review, and no accountability — and it reflects no one's convictions, including yours. Drawing on an authored argument and outsourcing your own voice are different acts, and the difference is exactly the part your congregation assumed you did.

This distinction isn't just theological instinct — it shows up in test data. In April 2026 we ran four AI tools through sixteen core doctrine questions and published the full results on our blog. On factual accuracy the tools were nearly tied: nine points separated first place from last. If AI sermon tools were unreliable at facts, this would be an easy article. They aren't.

The failures were elsewhere. One leading chatbot went zero-for-sixteen on pastoral appropriateness — not because it was heretical, but because every answer was hedged into 'some scholars argue' and 'there are different perspectives' framings that no one can preach. And only one of the four tools cited a verifiable source for any claim it made. Fluent, accurate-ish, source-less, and conviction-free: that is the raw material a generator hands you.

Which points to the real division of labor. AI is strong at gathering — surveying commentary positions, assembling historical background, listing cross-references, drafting discussion questions. It is weak precisely at asserting — the convicted, accountable, first-person truth-telling that preaching is. As leadership author Carey Nieuwhof has put it (paraphrasing him): AI makes a great research assistant and a dangerous preacher.

Here is a sermon-prep workflow that takes the gathering/asserting line seriously. Seven steps: Pray. Read. Reflect. Research. Review. Write. Pray again. The load-bearing rule: AI enters at step four — not step one.

Pray, read, reflect — the first three steps are you, the text, and God, with the laptop closed. Read the passage until it starts reading you. Note what confuses you, what convicts you, what you'd ask a scholar if one were sitting across the desk. This is the step a generator deletes, and it is the step the sermon actually comes from.

Research — now open the tools, because now you have real questions. Hand AI the gathering work: What do the major commentators say about this verb? What's the historical background of this city? Where else does this theme appear? What objections would a thoughtful skeptic raise? This is where AI genuinely earns its place — an hour of assembling sources compressed into minutes.

Review — treat everything it returned as unverified testimony (the next section gives you the checklist). Write — the manuscript is yours: your exegesis of your people's lives, in your voice, under your name. Use AI here at most as an editor asking whether a transition limps. Pray again — because 'unless the LORD builds the house, the builders labor in vain' (Psalm 127:1), and that is as true of sermons as of houses.

Five checks, every time, before anything AI-sourced reaches your manuscript. Are the references real? Verify every verse and every citation — AI systems demonstrably fabricate both. Does it align with your church's doctrine, or has it quietly blended traditions into a generic evangelical average? Is it pastorally usable, or hedged into academic mush? Are the scholars it quotes real people who actually said that? And the final gate: would you stake your name on this sentence, unedited, in front of the people who trust you? If a claim fails any check, it doesn't preach.

A simple spectrum answers most disclosure questions. AI as research assistant — surveying commentaries, checking background — is the moral equivalent of your commentary shelf: no disclosure needed, because the thinking and the words are still yours. AI as substantial drafter — outlines or paragraphs that survive into the manuscript — deserves one plain, proactive line: 'I used AI research tools in preparing this message.' And AI as ghostwriter — a generated sermon delivered as your own — isn't a disclosure problem; it's a different job than the one the congregation called you to.

One practical note: proactive beats discovered, every time. A congregation that hears you name your tools learns that you are trustworthy about your process. A congregation that finds out on its own learns the opposite, and it is very hard to un-learn.

Honesty about our own tool, since we make one. OpenLumin is deliberately not a sermon generator — we built it for step four, not step six. Ask it a research question and it answers with named scholars, dates, and passage references you can verify before you preach a word of it, and it's free. If what you want is a tool that writes the sermon, that isn't us, and after everything above you can guess why we made that choice.

Whatever tool you use, hold it to the standard this guide has argued for: it gathers, you assert; it cites, you verify; it serves the study, you do the preaching. The pulpit already has a preacher. What pastors need from AI is a research assistant that shows its work — and the discipline to keep it in that chair.