Using ChatGPT for Sermon Preparation: A Complete, Theologically Careful Guide (2026)

By Bobby Macintosh · Updated October 2026

Some weeks the sermon comes easily. Other weeks it’s Thursday night, the text still feels like a locked door, and someone at church has just suggested you “ask ChatGPT.” If that’s where you are, this guide is for you.

You wouldn’t be the only one asking. In Barna’s December 2025 survey of U.S. Protestant pastors, only 13 percent said they don’t use AI at all. Half use it to brainstorm, 36 percent use it to research biblical or theological topics, and 24 percent say they use it to write or edit sermons. Yet 71 percent also described themselves as “cautious” (Barna, June 2026). That mix of use and caution seems right to me, and it’s the spirit of what follows.

This is a practical guide. It walks through a week of preparation (study, exegesis, outline, illustration, application, editing) and gives you prompts you can copy and paste. It’s also a theological guide. The question isn’t only “Can AI help?” but “What kind of preacher am I becoming as I use it?”

First Things First: A Theology of the Study

Before any tool, there’s a calling. The apostles set their priorities plainly: “we will devote ourselves to prayer and to the ministry of the word” (Acts 6:4). Paul told Timothy to be “a worker who has no need to be ashamed, rightly handling the word of truth” (2 Timothy 2:15). When Ezra’s Levites read the Law, “they gave the sense, so that the people understood the reading” (Nehemiah 8:8).

None of that work can be handed to a machine: the praying, the careful handling of the text, the giving of the sense to a particular people. What can be handed over is some of the labour around it, like gathering background, testing structure, finding clearer words and catching mistakes. Keep that line clear and AI can be a real help in the study. Lose it and the tool starts shaping the preacher.

Church bodies across traditions have said much the same. The Southern Baptist Convention’s 2023 resolution urges believers to engage AI “from a place of eschatological hope rather than uncritical embrace or fearful rejection” (SBC, 2023). The Vatican’s January 2025 note Antiqua et Nova addresses pastors by name. It invites “those entrusted with transmitting the faith—including parents, teachers, pastors, and bishops” to give this subject careful attention, and it insists that generative AI “must be understood for what it is: a tool, not a person” (Antiqua et Nova, §§5, 59).

What AI Assistants Can and Cannot Do

What they do well

General-purpose assistants like OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini are good at language. They can:

Summarise and organise background material quickly.
Generate options such as angles, outline shapes, metaphors and questions, so you can choose rather than start from nothing.
Rephrase, simplifying jargon, tightening a paragraph or adjusting tone for teenagers or seniors.
Critique, pointing out gaps in logic, weak transitions, repetition and unclear sentences.

What they cannot do

They don’t know your text the way a lexicon or a careful commentary does, and they don’t know your people at all. Most importantly, they can be confidently wrong. OpenAI describes hallucinations as “plausible but false statements generated by language models,” and says plainly, “ChatGPT also hallucinates” (OpenAI, September 2025). For preachers, this is the main danger. A made-up verse reference, a misattributed quote from Augustine or Spurgeon, or a Greek “root meaning” that no lexicon supports can do real harm from a pulpit.

Christian-specific tools

Some Bible software now builds AI around its own libraries. Logos is the most developed example. According to Logos’s own help documentation (checked October 2026):

Study Assistant searches your Logos library and “composes a concise answer that cites your sources directly,” so you can check each claim against the book it came from.
Sermon Assistant (in Sermon Builder, on Logos Pro and Max subscriptions) has four modes: outlines, illustrations, applications and discussion questions. Logos itself notes that “the accuracy and usefulness of generated results should always be examined and adapted to your audience and personal experience.”
Logos says its Bible Word Study “relies exclusively on carefully curated proprietary data with no AI component.”
Logos says it currently uses models from OpenAI, Google, Microsoft and Anthropic, grounding them in curated content where it can.

(Sources: How Logos uses AI; What can I do with Sermon Assistant?)

My rule of thumb: use tools that cite checkable sources for facts. Use general assistants for form: structure, wording, options and critique. Never use either as the final word on what the text means.

A Seven-Stage Workflow for the Week

Stage 1: Pray and read the text yourself, first

Don’t open the chat window yet. Read the passage several times, in your own translation and at least one other. If you can, read it in Greek or Hebrew. Write down what you notice, what puzzles you and what convicts you. Pray it back to God. Carey Nieuwhof, interviewed by Barna, puts it well: “write the first draft yourself. Do the heavy lifting… pray, wrestle, fret… And then maybe go to AI in your second draft” (Barna, October 2026).

Your first observations are the most valuable part of the week. If AI speaks first, its framing tends to crowd out your own.

Stage 2: Background and context

Once you have your own questions, AI can give you a quick orientation. Treat it as a map you will check against better maps, not as the territory itself.

Prompt 1: Context briefing

I’m preaching on [passage] to a [describe congregation]. Give me a one-page briefing on the historical, cultural and literary context: author, audience, occasion, genre, and where this passage sits in the book’s argument. For each claim, say how confident scholars generally are, and flag anything that is debated. Do not invent quotations or page numbers. If you don’t know, say so.

Prompt 2: Interpretations across the tradition

Summarise the main ways Christians have interpreted [passage], including patristic, Catholic, Orthodox, Lutheran, Reformed, Wesleyan/Arminian and Anabaptist readings where they differ. For each, name the key interpretive move and the main textual reason given. Don’t tell me which is right. Then list the primary sources or major commentaries I should check to verify each summary.

The second prompt does two jobs. It exposes you to readings outside your tradition, and it gives you a reading list to check. Then actually check it: open the commentary, the Church Father or the confession itself.

Stage 3: Exegesis: structure, genre and original languages

This is where the tone has to get firmer. Exegesis is where AI is most tempting and most dangerous, because its mistakes sound like scholarship.

Language models produce plausible text. They don’t consult a lexicon unless a tool makes them, and even then you are trusting the tool. Etymological fallacies and invented “the Greek really means…” claims are common in human preaching too, and AI reproduces them fluently. Treat every lexical, grammatical or text-critical statement from a general assistant as unverified until you have checked it in a standard lexicon (such as BDAG for Greek or HALOT for Hebrew), a reference grammar and a good critical commentary. If you don’t read the languages, use an interlinear and trusted word-study tools rather than chatbot paraphrase.

Used in the right way, AI is a fine question-generator for exegesis:

Prompt 3: Literary structure

Here is [passage] in [translation]: [paste text]. Propose two or three possible outlines of its structure. Note repeated words, contrasts, chiasm, inclusio, parallelism or shifts in speaker/tense that support each outline. Quote the exact words from the text you are relying on, so I can check them.

Prompt 4: Word-study checklist, not word-study answers

I’m studying the word [Greek/Hebrew word, with transliteration] in [verse]. Don’t give me a definition. Instead, list the questions I should answer with a lexicon and concordance: its range of meaning, other uses by the same author, key Septuagint or Old Testament background, grammatical features in this verse, and common fallacies preachers fall into with this word.

Asking for questions keeps you doing the exegesis, which is where conviction is formed. As James warns, “we who teach will be judged with greater strictness” (James 3:1).

Stage 4: From exegesis to outline

By now you should be able to state the main idea of the passage in one sentence. Then AI can help you test the shape of the sermon.

Prompt 5: Three shapes for one message

The main idea of my sermon on [passage] is: “[your one-sentence big idea].” My supporting points from the text are: [list]. Suggest three different sermon structures: (1) expository, following the text’s order; (2) narrative/inductive; (3) problem–solution. Keep my points and my big idea. Don’t add new doctrine or new Scripture references without flagging them.

Prompt 6: Critique my outline

Here is my sermon outline: [paste]. Act as a careful homiletics professor. Where does the logic jump? Which point is weakest from the text itself? Where might a listener get lost? Does anything I say go beyond what the passage supports? Be direct, and be specific.

Prompt 6 is often more useful than asking AI to write an outline. You stay the author and the tool acts as a sparring partner.

Stage 5: Illustrations, handled honestly

Illustrations are where old “ChatGPT for sermons” advice most often went wrong. An assistant will happily invent a moving “true story” with names, dates and a tidy ending. If you tell it as true, you’ve told a falsehood from the pulpit. Ephesians 4:25 applies even to sermon illustrations: “having put away falsehood, let each one of you speak the truth with his neighbor.”

Prompt 7: Illustration ideas, clearly labelled

I need to illustrate this truth: “[one sentence],” for a congregation of [describe]. Give me six illustration ideas drawn from everyday life (work, family, nature, sport, ordinary objects). Clearly label each as a hypothetical or an analogy. Do not present anything as a true story, statistic or quotation. Avoid clichés and stereotypes.

Then reach for what only you have: what you’ve lived, read or been given permission to share. (For more, see Using ChatGPT to Illustrate Your Sermons.)

Stage 6: Application for real people

Here AI can widen your view. It can help you think about the single mum, the retired farmer and the student who all sit in the same pew. It can’t tell you which of those people are actually in front of you, or what they are carrying this month.

Prompt 8: Application across life stages

My sermon on [passage] calls listeners to [the response the text asks for]. Suggest concrete, specific applications for: a teenager, a young parent, someone in mid-career, a retiree, someone grieving, and someone exploring faith. Keep each one tied to the text, not to generic self-help. Note any place where an application might sound like law without grace.

If you preach across cultures, it’s also worth asking how a theme such as honour and shame, or debt and forgiveness, might be heard differently in your setting. Then check what you get against the people who know that culture from the inside.

Stage 7: Editing, clarity and delivery

This is where general assistants are most reliably useful, and least risky, because you are working on your own words.

Prompt 9: Plain language without losing truth

Rewrite this paragraph so a thoughtful 14-year-old could follow it, without losing or changing any theological claim: [paste]. Then list any terms you simplified, so I can check nothing was distorted.

Prompt 10: Tighten for the ear

This is a sermon manuscript meant to be heard, not read: [paste]. Suggest edits for the ear: shorter sentences, clearer transitions, repetition of the main idea at key moments, and cutting anything that doesn’t serve the big idea. Keep my voice; don’t make it sound polished or corporate.

Prompt 11: Discussion questions from my sermon

From the sermon text below, write eight small-group questions: two that check understanding of the passage, three that invite honest reflection, and three that move toward practice. Use only what’s in the sermon and the passage: [paste].

Prompt 12: Final verification audit

List every factual claim in this manuscript that should be checked before I preach it: every Scripture reference, quotation, historical claim, statistic, word meaning and named person. Present it as a checklist. Do not confirm them yourself.

Prompt 12 is my favourite: it turns the tool’s speed against its own weakness.

Writing Better Prompts

A few habits make every prompt above work better:

Give context. Name the passage, translation, tradition, setting (Sunday morning, funeral, youth night) and audience.
Assign a role. Asking for “a careful homiletics professor” or “an editor for spoken English” changes the response.
Ask for uncertainty. Tell it to flag debate and to say “I don’t know.”
Forbid fabrication. Say “no invented quotations, statistics or page numbers.”
Iterate. Treat the first answer as a draft. Push back and ask follow-up questions.

For more examples, see 7 ChatGPT Prompts to Deepen Your Sermon Research and Exegesis and Hermeneutics Prompts.

Limits and Cautions

This section is meant to be firm. These are not small matters.

Hallucinated verses, quotes and citations

Assume that any reference, quotation or citation from a general assistant is wrong until you’ve checked it. Antiqua et Nova names this exact danger, AI “hallucination,” where a system “yields results that appear real but are not,” and says everyone who uses these systems should “be committed to the truthfulness and accuracy of the information” they share (§86). Look up every verse in an actual Bible. Find every quotation in its original source, not just on a quotes website. If you can’t verify something, cut it.

Verifying against Scripture and the original languages

The Bereans “examin[ed] the Scriptures daily to see if these things were so” (Acts 17:11), and that was with an apostle. A chatbot deserves no less scrutiny. “Test everything; hold fast what is good” (1 Thessalonians 5:21). In practice: check claims about the original languages in lexicons and grammars, compare several translations, and read the passage in its whole context rather than in the snippet the AI quoted.

Theological bias and the “average” answer

Every model reflects its training data and its designers’ choices. The ERLC’s evangelical statement says AI “will be inherently subject to bias” (Article 5), and Antiqua et Nova notes that technological products “reflect the worldview of their developers, owners, users, and regulators” (§41). For preachers this has two sides. An assistant may quietly flatten distinctive convictions into a bland middle, or tilt toward whatever views dominate the internet on contested questions. Know your own confessional commitments, name them in your prompts, and notice when the answer drifts from them.

Plagiarism and honesty with your congregation

The prophets were rebuked for “steal[ing] my words from one another” (Jeremiah 23:30). Preaching another person’s sermon as your own was dishonest long before AI, and preaching a machine’s sermon as your own is no better. The SBC resolution asks Christians to use these tools “in honest, transparent, and Christlike ways… never seeking to willfully deceive others.” Antiqua et Nova states that “the use of AI should always be transparent and never misrepresented” (§84).

What does honesty look like in practice? Using AI to fix grammar probably needs no announcement. But if AI substantially shaped your structure, content or illustrations, be open about it. Pastor Sharon Hodde Miller, speaking at a Barna event, encouraged pastors to cite sources and name influences as a matter of integrity, saying “our integrity matters just as much as, if not more than, the outcome.” Carey Nieuwhof’s test is simple: “if you are uncomfortable disclosing how you use AI, you are overusing it” (Barna, October 2026).

Pastoral confidentiality and privacy

Never paste a congregant’s name, situation or confession into a general AI tool. Pastoral knowledge is held in trust. The ERLC statement affirms privacy as an individual right that should not be violated (Article 8). When you need help applying a text to a real situation, describe it generically.

AI does not replace prayer, the Spirit, or knowing your flock

This is the heart of the matter. Jesus promised that “the Spirit of truth… will guide you into all the truth” (John 16:13). Paul refused to rest his preaching on “plausible words of wisdom,” so that faith would rest “in the power of God” (1 Corinthians 2:1–5). A language model can give you plausible words faster than any preacher who ever lived. That is exactly why it can’t be the source.

Shepherds are told to “know well the condition of your flocks” (Proverbs 27:23), and to “shepherd the flock of God that is among you… being examples to the flock” (1 Peter 5:2–3). The Good Shepherd says, “I know my own and my own know me” (John 10:14). No tool has sat at a hospital bedside with your people, heard their doubts over coffee, or noticed who stopped coming. Your sermon should come out of that knowledge.

The ERLC statement denies “that humans can or should cede our moral accountability or responsibilities to any form of AI” (Article 3). Antiqua et Nova closes by reminding us that “we need the grace of the Holy Spirit” (§115). Pray before you prompt, and pray after.

Dependence and formation

Antiqua et Nova warns that users “should be careful not to become overly dependent on it for their decision-making” (§46). Wrestling with a hard text is not wasted time. It’s how God forms preachers. If you notice you can’t begin a sermon without the tool, take a few weeks off from it.

A Simple AI Policy for Your Preaching

Barna found that 94 percent of pastors think churches should have AI guidelines, but only 4 percent say their church has a formal policy and 17 percent informal guidance (Barna, October 2026). Here’s a starting point to talk over with your elders, vestry, deacons or parish council:

We pray and study the text ourselves before using AI.
AI may help with background, structure, wording and editing. It does not write our sermons.
Every Scripture reference, quotation, statistic and language claim is checked against original sources.
We never present AI-generated stories as true.
We never enter confidential pastoral information into AI tools.
When AI has substantially shaped a message, we’re willing to say so.

Frequently Asked Questions

Is it wrong for a pastor to use ChatGPT for sermon preparation?
Not in itself. Faithful preachers have long used commentaries and concordances. The real questions are whether you still do the praying and the exegesis, whether you check everything, and whether you are honest about how the sermon was made.

Can ChatGPT, Claude or Gemini write my sermon for me?
They can produce text that looks like a sermon. But a sermon is a pastor’s Spirit-dependent proclamation of a text to a particular people, and preaching AI-written text as your own raises serious honesty concerns. Use AI as an assistant, never as the author.

Is a Christian tool like Logos safer than a general chatbot?
For research, tools that cite the sources in your library make checking much easier. Even so, Logos itself says its generated ideas should be examined and adapted. No tool takes away your responsibility to check.

Should I tell my congregation I use AI?
If AI only helped with grammar or formatting, a note probably isn’t needed. If it substantially shaped the content, being open builds trust and keeps you accountable.

Can AI help with Greek and Hebrew?
It can help you frame good questions and find your way around tools. But lexical and grammatical claims must be checked in standard lexicons, grammars and commentaries. Don’t preach a word meaning you’ve only seen in a chatbot.

Sources and Further Reading

Barna Group, “Pastors Are Using AI More Than You Think” (June 15, 2026).
Barna Group, “Pastors Say Churches Need AI Guidelines. Few Have Them.” (October 6, 2026).
Dicastery for the Doctrine of the Faith & Dicastery for Culture and Education, Antiqua et Nova (January 28, 2025).
Ethics & Religious Liberty Commission, “Artificial Intelligence: An Evangelical Statement of Principles” (April 11, 2019).
Southern Baptist Convention, “On Artificial Intelligence and Emerging Technologies” (2023).
OpenAI, “Why Language Models Hallucinate” (September 5, 2025).
Logos Help Center, “How Logos Uses AI” and “What Can I Do with Sermon Assistant?“

Scripture quotations are from the ESV via Bible Gateway.

Keep reading on AI Prompt Theory: ChatGPT: Exploring Different Sermon Structures · AI-Generated Sermons: A Blessing or a Blasphemy? · The Ethical Considerations of AI in Biblical Interpretation · AI-Powered Biblical Greek Exegesis

3 views