7-Step AI Sermon Prep Workflow That Respects Text (2026)

If you’re a pastor reading this on a Sunday afternoon, you’ve probably done two things in the last 48 hours that almost no version of pastoral ministry from twenty years ago would recognize. You preached a sermon. And — depending on which Lifeway category you fall into — you either used a chatbot to help shape it, opened a chatbot once during the week and walked away unimpressed, or made a decision that you wouldn’t touch one with a ten-foot pole.

The numbers came out last week from Lifeway Research and Christianity Today, based on a phone survey of 1,003 US Protestant pastors conducted in September 2025 (±3.3% margin). 10% of pastors use AI regularly. 32% are experimenting. 18% are waiting to see better examples. 18% are intentionally avoiding it. 20% are ignoring it entirely. On the other side of the pew, 61% of churchgoers say they’re concerned about AI’s influence on Christianity, and 43% disagree with the idea of pastors using AI in sermon prep at all (24% strongly).

That’s the actual landscape. Not the hype version. There is no “all the pastors are using ChatGPT” — there’s a 42% of pastors touching it, a roughly even split among congregants on whether that’s even appropriate, and a real gap between the few pastors using AI well and the far larger group that tried it once on a generic prompt and got back something that read like a hospital chaplain’s first draft of a Hallmark card.

This guide is for the pastor who’s curious about the 10% — the regular users — and what specifically they’re doing differently. We’ll cover the seven-step workflow that respects the text, the four categories of grunt work AI is actually good at (and the three places it’s dangerous), the pre-Sunday voice-consistency check that closes the gap between “AI-assisted” and “AI-shipped,” and the honest reckoning with the church members in your pews who’ve already read the headlines and are wondering what you think.

Nothing here outsources proclamation. Proclamation is yours. We’re talking about cross-references, illustration brainstorming, transition smoothing, and the parent-comms email that always gets written at 11pm Saturday. The grunt work.

What the Numbers Actually Say (and Don’t)

Three things to set straight before any tool talk.

First, adoption is real but early. When 10% of pastors are regular AI users and another 32% are experimenting, that’s a 42% touch rate — meaningfully larger than the same number two years ago, but a long way from “everyone.” If you’re not using AI for sermon prep yet, you are not behind a curve. You’re sitting in the largest single bloc, with 58% of your colleagues either waiting for better examples, avoiding it on principle, or ignoring it.

Second, the demographic split is sharp. Lifeway’s data shows pastors 18-44 experimenting at 40% and pastors 65+ experimenting at 23% — a 17-point age gap. Doctorate holders are regular AI users at 14%; pastors without college degrees at 5%. Urban pastors run 11% regular vs. rural pastors at 5%. Holiness denomination pastors lead at 18% regular use; Lutherans and Baptists are most skeptical at 22% each ignoring AI entirely. If you’re a 50-year-old Baptist pastor in a town under 25,000, you are statistically the median pastor on this issue, and your colleagues’ reactions to AI sermon prep run the full range. That’s permission to take your time and arrive at your own answer.

Third, the pew is concerned in a specific way. That 61% churchgoer concern figure isn’t a vague “AI is scary” reaction. The more granular questions show people worry about errors (84% of pastors agree AI content needs heavy editing), unreliable sources (81%), programming bias (76%), and — the one that should haunt every pastor — lack of disclosure when AI was used (62%). The pastors who’ve gotten caught publicly using AI in the last six months — the Mar 26 viral TikTok, Carey Nieuwhof’s interview series, the comments under any Sean Cannell explainer — were almost never caught for using AI. They were caught for using it without telling anyone, and then having a generic AI line surface in front of a congregation that knew their voice well enough to spot the difference.

The honest read: pastoral AI use is fine; concealed AI use is a craft and trust problem. The seven steps below assume you’re going to be honest with yourself first about what work AI is doing, and then — sometimes — honest with your congregation about it.

Why Some Pastors Use AI Well and Others Get Caught

The pattern among the 10% regular users is not that they have better tools. They use the same ChatGPT or Claude that the 32% experimenters are using. The pattern is structural:

They give the AI a sustained set of instructions about who they are, what they preach, and which translation they prefer — once — and then reuse it every week, instead of starting from a blank prompt every time.
They use AI for the four bands of work where it’s measurably useful (cross-referencing, illustration brainstorming, transition smoothing, secondary comms) and never for translation choices, contextual judgment calls, theological framing, or the actual proclamation.
They run a voice-consistency check before Sunday — comparing the day’s draft against their own recent sermons to catch the “this doesn’t sound like you” moments before the congregation does.
They disclose in some lightweight, sustainable way — usually a one-line acknowledgment in the bulletin or on the church website, not a Sunday-morning preamble — so that congregants who notice phrasing shifts have a charitable interpretation already loaded.

That four-line list is the whole separation between “I use AI for sermon prep” landing well and landing badly. The seven-step workflow below operationalizes it.

The 7-Step Workflow

Block 45-60 minutes on a Tuesday morning. By the end of it, you’ll have your text exegeted at the AI-assistable level, an outline you wrote with reference material the AI surfaced, your bulletin and parent-comms drafted, and a voice-consistency pass complete. The two parts the AI never touches — your translation work and your final delivery — sit before and after this block, where they belong.

Step 1 — Set Up Your “Pastoral Assistant” Once (15 minutes, then never again)

In ChatGPT, this is a Custom GPT; in Claude, it’s a Project. Either works. The setup is the same: a one-time configuration that tells the AI who you are, so you stop repeating it in every prompt.

Paste the following template, with your details filled in. This is your “pastoral assistant” instruction set:

You are a research assistant for a working pastor. Below is who I am and how I preach.

DENOMINATION: [e.g., Southern Baptist, PCA, ELCA, non-denominational evangelical, Roman Catholic]
PRIMARY TRANSLATION: [e.g., ESV, NIV 2011, NRSVue, NRSV-CE, KJV]
PREACHING STYLE: [e.g., expository verse-by-verse, lectionary-driven thematic, narrative, redemptive-historical]
TYPICAL SERMON LENGTH: [e.g., 30 minutes, 45 minutes]
CONGREGATION CONTEXT: [e.g., 250-attendance suburban church, mostly families with school-age children, blue-collar economy, 3rd-generation members]
THEOLOGICAL FRAMEWORK: [e.g., Reformed covenantal, Wesleyan-Arminian, Anabaptist peace, classical Pentecostal, Roman Catholic post-conciliar]
WHAT I WANT FROM YOU:
- Surface cross-references, parallel passages, and patristic/Reformation/contemporary commentary worth my time
- Brainstorm modern illustrations that fit my context (no generic "imagine you're walking through a forest" stock examples)
- Smooth transitions in sermon drafts I write myself
- Draft secondary materials (bulletins, parent emails, small-group questions) once I've finalized my outline
WHAT I NEVER WANT YOU TO DO:
- Make translation choices for me
- Decide the theological framing of a text
- Write the actual proclamation portion of the sermon
- Tell me what the congregation should "feel"
- Use illustrations involving public figures, denomination-internal politics, or anything sourced after my last training-cutoff check
DISCLOSURE NOTE: Anything you draft will be revised by me. If you can't find a reliable source for a claim, say so explicitly rather than fill in.

Save it. Reuse it for every sermon for the rest of the year. The reason this matters is that the difference between “ChatGPT gave me Hallmark filler” and “Claude surfaced three cross-references I’d forgotten” is almost entirely in the upfront context. You wouldn’t ask a research assistant to help with sermon prep without telling them you’re a Lutheran from rural Wisconsin who preaches lectionary. Don’t do it with the AI either.

Step 2 — Run Exegetical Research on the Text (10 minutes)

With your assistant configured, paste the passage you’re preaching and ask for one specific category of help. Not “tell me about this passage” — that gets you the Hallmark version. Try:

Working from [your translation], surface the 5 most-cited cross-references for [Romans 8:18-30] across Reformed, Wesleyan, and patristic commentary traditions. Briefly note where they diverge in interpretation, especially on [the specific exegetical question on your mind]. Cite specific commentaries where you can; flag any claim you can’t source.

The “flag any claim you can’t source” line is the most important one in the prompt. AI hallucinates citations confidently — Augustine’s De Doctrina Christiana book IV chapter 7 may exist or may be invented, and the AI will say either with equal certainty. Your job is to verify the citations the AI gives you against your actual commentaries before any of them touch a pulpit. Treat it like a research assistant who’s smart but new — directionally useful, not authoritative.

Step 3 — Generate Outline Options + Illustration Brainstorm (10 minutes)

Same conversation, next prompt. Ask for outline shells, not finished outlines:

Based on the cross-references and themes we just discussed, sketch three different sermon-outline shells for a 35-minute message on [your text]: one expository (verse-by-verse), one thematic (one-question, three-supporting-points), one narrative (tension-resolution-application). For each, suggest two specific illustration directions that fit my congregation context (suburban families with school-age kids, blue-collar economy). Don’t write the illustrations themselves yet — just the directions.

The “don’t write the illustrations themselves yet” instruction matters because as soon as the AI writes the illustration, you’ll be tempted to use it. And generic AI illustrations are the #1 way pastors get caught — the “imagine a family driving home from a long vacation, the children asleep in the back seat” framing has appeared in roughly one-third of AI-assisted sermons since 2024. You want the AI to suggest that an illustration about family travel might fit here, then you find the actual story from your own life or your congregation’s that fits. Direction from AI, content from you.

Step 4 — Write the Sermon Yourself (the AI doesn’t touch this)

This step is one sentence long because it should be obvious and because it’s the line every “AI for pastors” post should make explicit:

You write the sermon. The AI does not.

You wrote sermons before AI existed. You will write sermons after this current cycle of AI hype settles into ordinary tooling. The proclamation is the thing your congregation actually came for, the thing that depends on your prayer life and your contextual judgment, and the thing that you will give an account for. The AI surfaced cross-references and outline directions. You take it from there.

Block 60-90 minutes for this. Phone off, pulpit Bible open, your own translation and your own thoughts.

Step 5 — Smooth Transitions (5 minutes)

Once you have a finished draft, you can ask the AI to do exactly one editorial job: smooth the rough transitions between sections. Not rewrite. Not restructure. Smooth.

Here is the full draft of my sermon. The proclamation, illustrations, and theological framing are mine and should not change. Read through, identify the 3-5 transitions between sections that feel rough, and suggest one-sentence fixes for each. Do not change anything else. Mark each suggestion with the line it replaces.

Then accept the suggestions you like and ignore the ones you don’t. This step takes minutes, but it’s the difference between a draft that flows and a draft that reads like seven separate notes stapled together.

Step 6 — Generate Secondary Comms in Parallel (10 minutes)

This is where the time savings actually live. Once your sermon outline is locked, the same AI conversation can produce — in one prompt — your weekly bulletin description, the parent-comms email summarizing what kids will hear in service, four small-group discussion questions for the following week, and a 90-second social-media teaser if you do that.

*Using the finalized sermon outline I just shared, draft the following four pieces. Each should match my preaching voice as established in earlier prompts:
A 60-word bulletin summary suitable for the printed program
A parent-comms email (200 words max) explaining what kids in grades 3-6 will hear in the family service this Sunday and one conversation prompt for the dinner table
Four small-group discussion questions for our Wednesday-night groups, calibrated for a mixed-experience adult group
A 90-second social-media post for Instagram (warm, not promotional)*

You’ll edit all four. But the time savings is real — what was 90 minutes of separate-format writing on Saturday night is now 15 minutes of editing on Tuesday afternoon.

Step 7 — The Pre-Sunday Voice Consistency Check (5 minutes)

This is the closing-the-gap step that the 10% regular users do and the 32% experimenters often skip. On Saturday night or early Sunday morning, paste your finished sermon back into your Pastoral Assistant with this prompt:

Here is my finished sermon for tomorrow. I’m also pasting the manuscript text from my last three sermons. Read all four together and flag any phrases, transitions, or illustrations in tomorrow’s sermon that feel inconsistent with my established preaching voice. Do not edit — only flag, with one sentence of explanation per flag.

The AI will surface 2-4 lines that read off-tone — a phrasing that’s a touch too modern, a transition that’s a touch too casual, an illustration that doesn’t quite fit your usual pattern. You decide whether to keep, edit, or remove each flag. This is the single best defense against the “this didn’t sound like you” reaction from your congregation, and it takes five minutes.

What AI Gets Dangerously Wrong About Scripture

The 10% who use AI well share a clear list of places they refuse to delegate. The three biggest:

One — theology hallucination. Ask an AI for a “Reformed take on sanctification” and you’ll get a fluent paragraph that sounds correct but mashes covenantal, Lutheran, and Wesleyan elements into a single position no actual Reformed theologian holds. The AI is pattern-matching on the word “Reformed,” not reasoning about it. Verify any theological framing claim against your own training and library before it lands in a sermon, full stop.

Two — citation fabrication. AI will invent commentary citations, denominational stats, and historical anecdotes with the same confidence it uses for accurate ones. When you ask “what did Calvin say about Romans 8:28,” you may get an accurate paraphrase, you may get a fabrication, and you cannot tell which by reading the AI’s output. If you don’t have the source on your shelf to verify, don’t preach it.

Three — denomination drift. Even with your “I am a [denomination] pastor” instruction loaded, AI tends to soften toward generic American evangelical phrasing over the course of a long sermon prep session. If you’re Lutheran, watch for missed sacramentology cues. If you’re Catholic, watch for generic “Bible-only” framing the AI substitutes for your tradition’s hermeneutic. If you’re Reformed, watch for synergistic language slipping in. The voice-consistency check in Step 7 catches a lot of this; periodic deeper review catches the rest.

The 18% of pastors who are intentionally avoiding AI in Lifeway’s data are not wrong about these dangers. They’ve decided the risk-to-benefit math doesn’t work for them, and that’s a defensible position. The 10% regular users have decided it does, and built containment around the failure modes. The pastors who get caught are the ones who didn’t build the containment.

What This Means for You

If you’ve never tried AI for sermon prep: Run Steps 1-3 once on a sermon you’ve already preached. Don’t use the output for anything; just see what the AI surfaces vs. what you ended up with on your own. Twenty minutes will tell you whether this tooling is going to be useful in your craft or not — and you’ll be able to give a more grounded answer to the parishioner who eventually asks you about it. We have a free 8-lesson AI for Church & Ministry course that walks through the same workflow with examples from real pastors who use it weekly.

If you’ve tried AI once, hated the output, and walked away: Steps 1 and 7 are probably what you were missing. Generic prompts get generic output; the upfront instruction set and the voice-consistency check are what move AI from “useless” to “useful for the grunt work.” Try the seven-step workflow on next Sunday’s text and see if the result is different.

If you’re already using AI weekly: The voice-consistency check in Step 7 is the single biggest upgrade most regular users haven’t installed yet. Five minutes on Saturday night closes most of the gap between “AI-assisted” and “AI-shipped.” If you’re not running it, install it this week.

If you’re a youth pastor: The bulletin / parent-comms / small-group-questions parallel-generation in Step 6 was almost certainly written for your week. You can take a Sunday-morning sermon manuscript, run a single follow-up prompt that turns it into a 4-week middle-school small-group curriculum with discussion questions, and ship it Tuesday morning instead of Friday at 11pm. We covered an early version of this workflow in the AI for Church & Ministry course and it’s the workflow that saves the most hours per week.

If you’re a parishioner who saw the Lifeway numbers and is wondering what your pastor is doing: Ask. Most pastors who use AI well will be glad you asked — the 62% pastor concern about non-disclosure is a real anxiety pastors carry, and a charitable conversation gives them room to explain. Most pastors who don’t use AI will also be glad you asked — they have thought about it and have a defensible answer either way. The conversation is the thing.

The Bottom Line

The Lifeway and Christianity Today numbers landed last week, and the actual story is less dramatic and more interesting than the hype version. Pastoral AI adoption is real but early. Congregant concern is real and specific. The pastors using AI well are doing four structural things — sustained instructions, narrow scope, voice-consistency checks, lightweight disclosure — that the pastors getting caught are not.

If you’re going to use AI for sermon prep, the workflow above is a defensible craft-respecting version of it. It does not replace the work of a sermon. It does, on a good week, give you back two or three hours that you spend on things only you can do — pastoral visits, prayer, the slow re-reading of a passage that isn’t billable to any spreadsheet.

That, more than the cross-references or the bulletin draft, is the actual point.

Sources:

The 7-Step AI Sermon Prep Workflow That Respects the Text

Table of Contents