What counts as AI-assisted preaching?

A disagreement worth making precise

A church member who asks whether Sunday's sermon used AI deserves an answer more informative than yes or no. A pastor might have located a commentary, accepted an interpretation, reorganized an argument or delivered generated paragraphs. Each use can affect the message, but each places a different responsibility on the preacher. The useful question is what judgment or expression the tool supplied, and what the pastor did with it.

Two recent Christian essays sharpen that question. In an October 1 Christianity Today review of Carl Trueman's Classical Theology for the Modern Mind, pastor Claude Atcho connects AI-assisted ministry with a deeper preference for expediency and therapeutic reassurance. His argument concerns the church's theological habits, rather than offering a technical test of sermon software. Source: christianitytoday.com

Writing in Church Executive on September 29, theologian David Kotter makes room for administrative and research assistance while insisting that pastors retain prayer, scriptural engagement and the work of knowing their people. His positive examples do not demonstrate that every suggested use performs reliably. Source: churchexecutive.com

Both essays press beyond whether a tool saves time. Congregations now need language precise enough to examine the work that time-saving changes.

What the survey does and does not count

Barna's June 15 report provides historical context, not a new October adoption count. Its pastor findings came from an online survey of 442 U.S. Protestant pastors in December 2025, using quotas for church size, denomination and region. The research series is a partnership with Gloo. Barna reports that 24% used AI to write or edit sermons and 36% used it for biblical or theological research. Source: barna.com

Those categories leave consequential questions unanswered. Writing and editing are combined; the reported figure does not tell us how much material was generated, how often it was used, or how pastors checked it. Biblical research is broader than sermon preparation.

The article also compares the 24% practice figure with 12% who expressed comfort with AI sermon-writing in earlier research. Comfort and behavior are different measures. Those numbers cannot establish that sermon-writing adoption doubled.

Lifeway's separate pastor survey asked respondents to select their current relationship with AI for ministry, finding 32% experimenting and 10% regular users. Its 1,003 telephone interviews ran September 2–24, 2025, with a stratified church sample, church-size quotas and weighting by region and church size. That different question and method cannot be combined with Barna's task categories into an adoption trend. Source: research.lifeway.com

Follow the contribution through the sermon

For this analysis, AI-assisted sermon preparation means using generative tools to contribute to the message's research, interpretation, composition or revision. That working definition is an editorial aid, not a category established by either survey. It helps separate four kinds of contribution.

First comes finding material. A tool may suggest a commentary or locate a passage. The pastor still needs to open the source and decide whether it supports the point. Receiving a summary without reading the underlying argument moves beyond finding into interpretation, even when the interface calls everything research.

Second comes shaping the argument. Asking for an outline or a central claim can delegate more than its small word count suggests. An outline determines emphasis, sequence and what receives no attention. A preacher who rewrites every sentence may still be following a generated interpretation. Counting retained AI words would miss that dependence.

Third comes expression. Correcting punctuation differs from supplying an analogy, a conclusion or an appeal to repentance. An editorial suggestion can clarify something the pastor already understands; a generated paragraph can also conceal an argument the pastor has not examined. A useful self-test is whether the preacher can explain why each important claim belongs in the sermon without asking the tool again.

Fourth comes material published afterward. A generated summary or discussion guide should be checked against the sermon actually preached. If it introduces an application the preacher never made, it has become additional teaching. Calling it repackaging does not remove the need to approve it.

These distinctions make room for practical assistance while showing why “I only used it for preparation” is an incomplete account. Preparation is precisely where a sermon acquires its meaning.

Congregations are divided over an undefined kind of help

Lifeway's churchgoer survey illustrates the difficulty. Among 1,200 American Protestant churchgoers, 44% agreed that they saw nothing wrong with a pastor using AI to help prepare a sermon; 43% disagreed and 13% were unsure. Respondents came from a pre-recruited online panel surveyed September 2–7, 2025. They identified as Protestant or nondenominational and attended services at least monthly. Quotas and weights addressed demographic and religious characteristics. Source: research.lifeway.com

The question did not present separate scenarios for proofreading, generated exegesis or a complete manuscript. It therefore cannot tell a church which of those practices its members would accept. Nor is a national response a vote by one's own congregation.

Pastors also recognize a disclosure issue: 62% in Lifeway's separate survey selected concern about users failing to disclose AI collaboration. That measures concern, not the prevalence of concealed use. Source: research.lifeway.com

A congregation can improve the conversation by discussing actual categories before a controversy. Leaders might describe two fictional preparations, one involving checked source suggestions and another a generated outline and conclusion. Ask what listeners believe the preacher is taking responsibility for in each. Disagreement may reveal different expectations of preaching, not simply different enthusiasm for technology.

Christian perspective: preparation forms the preacher

Second Timothy 3:14–4:5 places preaching within a larger account of learning, Scripture and ministry. Timothy is reminded of the people from whom he learned and of Scripture's relation to salvation through faith in Christ. Its teaching and correction equip a person for good work. The charge to preach then includes patience, careful instruction and perseverance when hearers prefer more agreeable teaching. Read the passage in the NIV: 2 Timothy 3:14-4:5 (NIV)

The passage offers no software policy. Its challenge is that faithful teaching cannot be judged solely by its welcome. Applied here, a tool's contribution requires examination against Scripture, including when that means abandoning an attractive outline.

There is also a question about the preacher's formation. The work of being corrected by Scripture can alter what a minister intends to say. A process that begins with a desired conclusion and asks only for supporting religious language risks excluding that encounter. This is a temptation in human sermon preparation too; using no AI is no guarantee of faithfulness.

That is why personal responsibility means more than approving fluent prose. The preacher needs sufficient understanding to teach, receive challenge and correct a mistake. Assistance is most defensible when it strengthens those capacities. Churches can ask that question despite differing tool policies.

Check the substance and the implied witness

The technical reason for verification is well established. NIST's July 2024 Generative AI Profile describes confidently presented false content, including invented citations or reasoning, as a risk of generative systems. It recommends checking sources and citations in outputs. This is general risk guidance, not evidence of an error rate for sermons or a verdict on a particular current product. See section 2.2 and action MS-2.5-003. Source: nvlpubs.nist.gov

For sermon preparation, the practical response is to trace important claims to material the preacher has actually inspected. Read biblical quotations in context and confirm their translation. Check a historical anecdote at its source. Treat an attractive original-language explanation as a question for a reliable reference or a competent reader, rather than evidence of expertise merely because Greek or Hebrew appears on screen.

A separate check concerns what the speaker appears to have experienced. A generated first-person illustration must not be presented as the pastor's memory. A hypothetical story can be identified as hypothetical. Disclosure that AI helped somewhere in preparation would not make a false personal testimony truthful.

These checks also protect the member who asks a sincere question afterward. The preacher should be able to show the basis of the teaching and explain its limits, rather than returning the question to an opaque preparation process.

A disclosure that helps a listener decide

AI Faith Monitor's recommendation is to disclose consequential contributions, with detail proportionate to their role. This is a proposed congregational practice, not an existing universal church rule. A standing explanation may adequately cover routine proofreading. A generated argument, illustration or substantial passage calls for a sermon-specific account accessible to its listeners.

For example, if accurate, a preacher could say: “I used an AI tool to suggest an initial outline. I checked its claims against the sources listed in the sermon notes, revised the structure and take responsibility for the teaching.” If generated wording remains, that should be said too. “AI-assisted” alone does not explain that distinction to the listener.

Disclosure does not resolve every concern. A congregation may decide that accepting a generated central interpretation is incompatible with its understanding of preaching, even when announced. Another may permit bounded drafting. Either needs an agreed boundary, a responsible church leader to hear questions and a way to correct published material.

Pastors should also be able to raise workload problems without feeling required to demonstrate greater technological productivity. If sermon preparation is repeatedly being squeezed out, leaders should examine expectations and support, rather than treating a faster manuscript as proof that the underlying problem is solved.

The next useful conversation is therefore specific: what did the tool contribute, how was it tested, and who will answer for it? A congregation does not need access to every working manuscript. It does need an honest account of the teaching offered in its name.