Should pastors use AI at all?

For mechanical work, most thoughtful evangelical voices say yes. Transcribing a recorded sermon, drafting captions, organizing files or a volunteer schedule are tasks a machine can do without touching anyone's soul. The consistent warning, from writers at 9Marks, The Gospel Coalition and elsewhere, is against handing over spiritual work: writing the sermon, counseling the hurting, praying with the dying.

Is it wrong for a pastor to use AI to write a sermon?

Many serious pastors and theologians say yes, or close to it. Trevin Wax at The Gospel Coalition argues that relying on a machine to prepare your sermon removes 'the heart' of preaching, and relays John Piper's view that it is 'wicked to rely on a robot to generate the first draft of a sermon.' The preparation itself, wrestling with the text in prayer, is part of what forms the preacher.

What are AI tools actually good for in a church?

The mechanical layer of ministry: turning sermon recordings into transcripts, generating captions, producing a first translation for a fluent member to review, organizing files and scheduling. In each case a person decides what the church says; the machine handles the typing, the timing or the conversion between formats.

Is sermon transcription the same as AI writing sermons?

No, it's the opposite operation. Speech recognition writes down the words it hears, in the order they were said: the input is what the pastor preached and the output is those same words as text. A sermon generator produces content no pastor wrote. The two get lumped together as 'AI,' which is why churches should ask what a tool does, not what it's called.

Does our church need an AI policy?

It helps more than you'd think, because the line between administrative and pastoral work blurs in practice. Writing down, as a leadership team, which tasks you'll automate, which you never will, and who reviews machine output before it's published turns a fuzzy instinct into a shared standard.

Guide

AI for Churches: What It Should (and Shouldn't) Do

By the SermonKeep team · October 9, 2026

Ask ten pastors about AI for churches and you’ll hear everything from “it saved our media team ten hours a week” to “it has no place in the house of God.” Both reactions are usually about different things wearing the same label. “AI” now covers tools that type up what a preacher said and tools that claim to write what a preacher should say, and a church that doesn’t tell them apart will either adopt something it should have refused or refuse something that would have served its people.

We build one of these tools, so let’s be upfront: SermonKeep is our sermon transcription service for churches. That gives us an interest in the question, and it has also forced us to think hard about where the line sits. This post is our attempt to draw it plainly, with help from people wiser than us who have written on exactly this.

AI for churches: one line that makes the decisions easier

Here’s the framework in a sentence: AI belongs in the mechanical work of the church and has no place in the spiritual work of the shepherd.

Mechanical work is anything where the content already exists or the task is procedural: turning a recording into text, formatting a bulletin, matching volunteers to slots, answering “what time is the Christmas Eve service?” Nobody’s soul is at stake in the typing. Spiritual work is where a person made in God’s image meets the Word of God or another person: preaching, counseling, prayer, discipleship, sitting with the grieving. That work is the ministry; a machine doing it isn’t a shortcut, it’s a substitution.

Two vocabulary notes, because plain names beat mystique:

Speech recognition is a model trained on recorded speech to do one job: write down the words it hears, in the order they were said.
A language model (the technology behind ChatGPT and similar tools) predicts likely next words, and can draft, summarize and rephrase text.

Neither understands, believes or prays. Keeping the names plain keeps the decisions clear.

What AI should do in a church

Applying the line, here’s where these tools genuinely serve a congregation:

Sermon transcription. Speech recognition turns Sunday’s recording into text you can read, share and look back through. It’s the highest-leverage use we know, because accessibility, study material and a sermon library all flow from it. Our page on sermon transcription for churches explains how it works.
Captions and subtitles. Automatic captions make your livestream and clips accessible to members who are Deaf or hard of hearing. A volunteer reviews; the machine does the timestamping.
Translation drafts. For bilingual congregations, a machine-translated first pass of a transcript or bulletin, reviewed by a fluent person before anything is published, turns hours into minutes.
Administrative work. Gathering form responses, organizing files, the fourth reminder about the parking lot, the first draft of a volunteer rota. A person still hits send.

Notice the shape every good use shares: a person decided what the church says; the machine handled conversion, formatting or logistics; and a person checked the output. No transcript, caption file or translation should go out unreviewed. Speech recognition still mishears theological vocabulary, and a caption that garbles a Scripture quotation is a mechanical task done badly.

What AI shouldn’t do

The other side of the line, just as plainly:

Writing sermons. Not the outline, not the first draft. The sermon is the fruit of a called shepherd wrestling with the text in prayer for these people, and that wrestling forms the pastor, not just the manuscript.
Pastoral counseling. A language model produces plausible, agreeable text. It can’t know a person, carry their history, or answer before God for what it says.
Pastoral care and prayer. A machine-generated “I’m praying for you” is a false sentence. Nobody is praying.
Answering members’ spiritual questions in the church’s name. A chatbot on your website confidently misstating your church’s teaching is a real and well-documented failure of language models.

In each case the machine wouldn’t be assisting the shepherd; it would be standing in for the shepherd. And a congregation can tell. The trust a church gives its pastor’s words collapses the moment those words turn out to be nobody’s.

There’s an accountability question underneath, too. A pastor answers before God for what they teach and how they care for the flock. When a language model drafts the counsel or the sermon, who answers for it? That question has no good answer, which is itself the answer.

Should pastors use AI? What trusted voices say

We didn’t invent this line. It’s close to a consensus among evangelicals who have written carefully on the subject. A few pieces worth reading in full:

Trevin Wax, in “The Missing Heart in AI-Generated Sermons” at The Gospel Coalition (April 2025), argues that “relying on a robot to prepare your sermon is to excise an indispensable element of good preaching—the heart.” He relays John Piper’s stronger view, that it is “wicked to rely on a robot to generate the first draft of a sermon,” because worship is “right feeling about God,” and only people can adore.

Benjamin Eggen, a pastor in Switzerland writing at 9Marks, asks “Are AI’s Promises of Productivity Worth the Risks for Pastors?” (July 2025). He grants that AI can handle real administration, like gathering form data and organizing files, then presses the uncomfortable question: “the distinction between administrative work and pastoral work is not always clear.” His summary is the one we’d put on a poster: “AI might make you more efficient, but it cannot make you godlier.”

Willy Kotiuga, on the Lausanne Movement’s blog in “Artificial Intelligence: Servant, Not Substitute” (November 2025), frames it as stewardship: AI can help draft and organize, but “it must never be allowed to substitute for human discernment, relational intelligence, or spiritual authority.”

On the EFCA blog, “What Should Pastors Do About Artificial Intelligence?” (December 2025) pairs two reviews of Todd Korpi’s AI Goes to Church. Lowrie Robertson, a missionary educator and church planter, urges pastors to “use AI to streamline the pastoral and administrative workflow to free up more time for face-to-face ministry.” Cody Crumrine, a software engineer and church elder, warns that “pastoral AI” will be “more convincing than accurate.”

And Ed Stetzer, on Biola University’s Good Book Blog in “Navigating the Ethics of AI in Ministry and Sermon Writing” (October 2025), offers the most memorable image: AI as the sous chef in the ministry kitchen, while “the pastor remains the ‘head chef’ responsible for vision and execution.”

Different traditions, one pattern: mechanical help, yes; substituted shepherding, never.

Transcription is not generation

This is the distinction that gets lost most often, so it gets its own section. When headlines talk about “AI sermons,” they mean generation: a language model producing sermon content no pastor wrote. Sermon transcription is the opposite operation. The input is what your pastor actually preached; the output is those same words as text. Nothing spiritual is created, because everything spiritual already happened in the study and the pulpit.

Here’s how that line works in our own product, stated plainly because you should be able to check it:

Speech recognition (AssemblyAI’s model) writes down the sermon, word for word.
A language model then does three kinds of work, and only on that transcript: it proposes fixes for words that look misheard (obvious misrecognitions are fixed and listed so you can undo them; anything uncertain waits for you), it finds the Scripture the sermon used, and, when you ask, it drafts material from the sermon: a summary, a small group guide, clips and quotes, a YouTube description.
It never writes a sermon, and the material it drafts is built only from the preacher’s words. The small group guide quotes the preacher and points to the minute each answer was said; it uses only Scripture the preacher cited and doesn’t settle questions the preacher left open. Quotes are never rewritten.

When you evaluate any church AI tool, ask one question first: does this tool produce words on someone’s behalf, or process words someone already said? Processing is a photocopier. Producing on someone’s behalf is a ghostwriter, and the pulpit doesn’t take ghostwriters. (If you’re weighing speech recognition against a human transcriber, that’s a question of accuracy and cost, not theology; we compare them in human vs automatic transcription.)

Put the line in writing

One honest complication: the line is clear at the extremes and fuzzy in the middle. A reply to a scheduling email is plainly mechanical; a reply to a member asking why God allowed her miscarriage is plainly pastoral; a reply to “can you recommend a book on parenting?” sits somewhere between. Eggen is right that the boundary blurs in practice, which is why every church that adopts these tools should decide the gray cases together, in advance, rather than leaving each staff member and volunteer to improvise. We’ve put together a church AI policy template you can adapt: which tasks you’ll automate, which you never will, and who reviews machine output.

A written policy does one more quiet job: it protects trust. If a member ever asks “did a computer write this?”, the answer should never be a shrug. A church that can say “here’s exactly what we use machines for, and here’s what we never will” has nothing to hide.

The bottom line

AI in ministry isn’t one question; it’s two. Should a machine do the church’s typing, timestamping and scheduling? Gladly; that’s time returned to people. Should a machine do the church’s preaching, praying and shepherding? Never; that’s the ministry itself, and it belongs to people called by God to other people.

If the mechanical side is where your church could use help, transcription is the natural first step, and we’ve tried to build ours on the right side of the line. SermonKeep turns your pastor’s recorded preaching into a faithful, word-for-word transcript, with the Scripture found and material for the week drafted only from what was preached. Your first sermon is free, up to an hour, no card required. The words stay your pastor’s. That’s the whole point.