Generative artificial intelligence is becoming part of sermon preparation in American churches, moving from administrative tasks into one of a pastor’s most consequential responsibilities.

Clergy are using chatbots to examine biblical passages, test arguments, develop outlines, find illustrations and refine prose.

A smaller but growing group is allowing the technology to draft or edit portions of the sermon itself.

Darrell Stetler II, lead pastor of Bible Methodist Church in Oklahoma City, embodies that tension.

About 20 hours before a recent Sunday service, he worked in the church sound booth with presentation slides on one monitor and ChatGPT on the other.

His sermon, titled "Forged," belonged to a summer series on the Book of Proverbs and emerged from his own notes, research and exchanges with several artificial-intelligence systems.

Stetler, 45, is an unusually intensive user.

He is among ChatGPT’s most active 1 percent of customers, administers an online group of approximately 500 pastors experimenting with artificial intelligence and created a tool that can generate numerous possible sermon illustrations within seconds.

Yet he rejects the idea that computational fluency amounts to belief, prayer or spiritual understanding.

That distinction became vivid when a chatbot ended one exchange by telling him it was praying for his preaching.

Stetler’s immediate reaction was that the system was doing no such thing.

The sentence resembled pastoral encouragement because the model had learned its linguistic pattern; it did not represent an act of prayer or an intention held by the machine.

For Stetler, the practical attraction is time.

Pastors, particularly those in small congregations, may be responsible for preaching, counseling, hospital visits, administration, volunteer management and family care without a research or communications staff.

Automating preliminary research or generating examples can preserve hours for work that requires physical presence and a personal relationship.

Adoption is rising, although survey results differ because researchers define artificial-intelligence use and sermon preparation in different ways.

One representative survey of 442 Protestant pastors found that 87 percent used artificial intelligence somewhere in ministry.

Half employed it for brainstorming, 36 percent for biblical or theological research and 24 percent for writing or editing sermons.

That last figure was twice the 12 percent who said in early 2024 that they were comfortable using artificial intelligence to write sermons, although the two questions measured practice and acceptance rather than precisely the same behavior.

A separate telephone survey of 1,003 Protestant pastors produced a more conservative estimate.

It found that 42 percent were either experimenting with or regularly using artificial intelligence for ministry, while 56 percent were waiting, deliberately avoiding it or ignoring it.

Younger pastors, ministers at larger churches and urban clergy were generally more likely to experiment than older, rural or small-church pastors.

The surveys nevertheless identify similar anxieties.

Eighty-four percent of pastors in the larger study said artificial-intelligence output must be edited on the assumption that it contains errors, while 81 percent questioned whether the tools rely on trustworthy material.

Seventy-six percent worried about embedded bias, 62 percent raised concerns about undisclosed machine collaboration and 59 percent regarded the reuse of material from other sources as a potential form of plagiarism.

The theological objection goes beyond factual reliability.

In many Christian traditions, sermon preparation is itself considered a spiritual discipline: the preacher studies scripture, reflects on the circumstances of a particular congregation and develops a message through prayer, experience and moral judgment.

A chatbot can reproduce the vocabulary of that process but cannot participate in it as a conscious believer.

The concern is not that every technological aid disqualifies a sermon.

Pastors have long consulted commentaries, borrowed structures, searched digital concordances and used spelling or presentation software.

The emerging argument concerns degree and authorship.

Using a model to locate cross-references is closer to research assistance; delivering a largely machine-generated manuscript as personal pastoral reflection raises a different question about honesty and vocation.

Some clergy have ventured considerably further.

Justin Lester, a pastor at Friendship Baptist Church in Vallejo, California, created PastorGPT, a system that can be trained on a minister’s previous sermons and teaching.

It can turn sermon manuscripts into study material, church communications and other resources while attempting to preserve the pastor’s theological and rhetorical patterns.

Lester says versions of the system are operating in more than 30 churches.

His broader experiment includes Theo, a robotic dog associated with the project.

PastorGPT can also analyze anonymized interactions for recurring subjects raised by users, allowing a church to identify concerns that may warrant a lesson or pastoral response.

Lester has said that an increase in conversations about teenage bullying prompted him to address the issue.

That function introduces a separate set of risks.

Even anonymized pastoral conversations may contain sensitive information about mental health, family conflict, sexuality, abuse or faith.

Churches using such systems must determine where data are stored, who can examine them, whether users understand that they are communicating with software and how urgent disclosures involving self-harm or violence will reach a qualified person.

The Reverend Dwight Lee Wolter of the Congregational Church of Patchogue in New York has used artificial intelligence more directly in worship.

After the attempted assassination of Donald Trump at a Pennsylvania rally in 2024, Wolter asked ChatGPT what he should tell his congregation the following morning.

He judged the resulting prayer suitable and has also used artificial intelligence for calls to worship, prayers and newly composed hymns with generated music.

Wolter maintains that he retained responsibility rather than surrendering the service to a machine.

The dispute therefore cannot be reduced to a simple division between technology-friendly and technology-resistant churches.

Many of the most active users are also among the most explicit about artificial intelligence’s limitations.

The practical boundary commonly proposed is that software may assist with research, organization and revision, but the pastor remains accountable for every factual assertion, theological claim and word delivered to the congregation.

Disclosure is less settled.

Some church guidance permits artificial-intelligence content when its use is attributed, while other denominational statements offer broad principles about human dignity and discernment without specific rules for sermons.

Most congregations still lack a formal policy explaining whether pastors must disclose machine assistance or how substantial that assistance can become before attribution is expected.

The absence of common standards leaves individual ministers to decide when assistance becomes substitution.

For Stetler, the warning lies in the possibility of producing religious language without the inward engagement it purports to convey.

He continues to use artificial intelligence, but treats its output as material to interrogate rather than spiritual authority, retaining personal responsibility for the sermon preached from his pulpit.