AI for Church Leaders Stops at the Pulpit

Church Leaders Let AI Run the Office, Not the Pulpit

Eighty-seven percent of U.S. Protestant pastors now use AI in their ministry, according to research from Barna and Gloo released in mid-June. Only a shrinking minority avoid it entirely. For a profession often described as cautious about new tools, that figure quietly closes a two-year argument over whether church leaders should touch this technology at all.

The more useful finding sits under the headline. The same pastors who reach for AI every week are keeping it away from the parts of the job they consider sacred. They let it brainstorm a sermon series, gather commentary, lay out a slide deck, and clear the scheduling backlog. They do not let it preach, sit with a grieving family, or speak for God. What church leaders are deciding now has moved from permission to placement: which work an agent may take on, and which work stays in human hands.

The adoption argument answered itself

The newest data on how churches are using AI comes from the 2026 wave of Barna and Gloo's research on faith and technology, a survey of Protestant pastors fielded late last year. Its headline finding, near-universal adoption, matters less for its scale than for where the usage actually lands. Pastors are not running AI at the front of the sanctuary. They are running it in the back office.

Cluster the reported uses and the shape is obvious. The common applications are preparation and production: generating ideas, researching a passage, drafting discussion questions for a small group, producing graphics, and handling administrative work like scheduling and email. None of those touch the moment a pastor stands up to speak or sits down to counsel. The number of churches using AI for that routine work keeps climbing, while the sacred functions stay off-limits.

That shift changes the decision in front of church leadership. For two years the conversation about AI and the church ran as a yes-or-no vote. It has become an operations question, and most congregations are answering it without a map. Adoption is widespread; written guidance is rare. Very few churches operate with a written church AI policy, which means the boundaries pastors clearly feel are mostly held in their heads rather than on paper. For most leaders the harder task is the one that comes after the yes: writing down where these tools are allowed to act, and where they are not.

In practice that looks less dramatic than the headline suggests. It means naming which tasks an assistant may handle on its own, which ones a staff member reviews before anyone sees them, and which ones never leave a human's hands at all. A church that has agreed AI can format the bulletin but cannot answer a member's late-night message about a death in the family has already done the hard part. It has identified the work that carries weight and protected it, and it has freed everything else to move faster.

Prep yes, proclamation no

The sermon is where the line shows up most clearly. AI use for sermon preparation has roughly doubled over the past two years, but pastors describe it as a research partner rather than an author. It assembles background, surfaces commentary, and suggests an outline. The writing, the conviction, and the act of delivery stay the pastor's own. The tool is welcome up to the edge of the pulpit and stops there.

The workflow most pastors describe is mundane and revealing. They ask a model to pull together what several commentaries say about a passage, to suggest a structure, or to test whether an illustration lands. Then they throw most of it out and write the thing themselves, because a sermon works only when this person, to this congregation, is willing to stand behind the words. AI shortens the gathering. The claim still has to be made by someone who means it.

That instinct is not confined to one tradition. In the same stretch this month, Catholic clergy in one U.S. diocese said plainly that they would not hand a homily to a machine, echoing guidance that authentic preaching has to come from a person who has actually wrestled with the text. Protestant and Catholic leaders are drawing the same boundary from different starting points, which suggests it reflects something about the work itself rather than the preference of any single denomination.

Read plainly, the boundary is not technophobia. The questions pastors are raising about artificial intelligence and the church concern authorship and accountability: who is responsible for the words, and whether a congregation can trust that a real person did the work of preparing them. A leader will happily let AI collect the raw material. The claim that follows, the part a community is asked to believe, has to belong to a human who can stand behind it.

Why the line falls where it does

The reasoning becomes clearer once you look at what these tools do well and where they fail. A language model is fluent by design. It produces confident, well-formed prose whether or not the underlying claim is correct, and in a sermon a fluent error is dangerous precisely because it sounds authoritative. A congregation cannot easily check a citation to Scripture delivered with conviction from the front of the room, and a misquoted verse or an invented source carries further when it arrives inside a trusted voice.

So pastors are sorting AI into two roles. As a research aide it is useful and safe, because a person checks its output before anything reaches the congregation. As an authority it is unacceptable, because the claims it would make are exactly the ones a congregation is asked to take on faith, with little practical way to verify them in the moment. Seventy-nine percent of pastors in the research say they worry the technology could stand in for God in someone's spiritual life, and that concern is less about efficiency than about what gets lost when a relationship built on a real human presence is handed to software.

Authenticity is doing quiet work underneath that worry. A congregation's trust runs partly on the assumption that the person speaking has actually sat with the text, the loss, or the decision in front of them. The effort is part of the message. A leader who outsources the thinking and keeps only the delivery spends down a kind of trust that is hard to earn back, even when the words themselves are good.

This is also the deeper issue under the broader conversation about artificial intelligence and Christianity: which acts of faith are supposed to be performed by a person, regardless of what a tool can technically do. The distinction pastors are enforcing, verifiable help on one side and unaudited authority on the other, is a useful one well past the church doors. In any field where fluent output can be confidently wrong, the safe pattern is AI that surfaces material a human checks, not AI that issues conclusions nobody reviews.

A boundary any AI adopter has to draw

Strip away the theology and church leaders are describing a control model that applies far beyond the sanctuary. Let AI handle preparation and back-office work; keep a human owning anything sensitive, irreversible, or trust-bearing. A church that lets software draft the newsletter, translate the service, schedule volunteers, and reconcile the week's giving, while keeping a person on anything sacred, is doing more than expressing a preference. It is describing how it wants an automated system to behave. Translate that into an operating rule and it reads like the one any company deploying AI agents eventually writes: an agent can do the work, every action it takes is logged, and a person signs off before anything consequential goes out the door. The categories shift by industry; a missed homily and a mistaken wire transfer sit at very different stakes. The structure underneath stays identical: name the work that must remain human, automate the rest, and keep a record of everything the system does in between.

That is the line AgentPMT is built to make enforceable. As model-agnostic orchestration for AI agents, it lets a team choose which agents and tools their work runs on, records every action an agent takes, down to the full request and response, in an audit feed, and requires a human approval before an agent does anything expensive or sensitive. Budgets, per-tool limits, and an encrypted vault for credentials keep an agent inside its lane without anyone handing over the keys. None of that lets software do a pastor's job, and it should not try. What it does is take the boundary pastors draw by instinct, AI in the workroom and a human at the pulpit, and let any operator set the same boundary on purpose, with the system holding the line rather than a policy memo hoping it holds.

The faith sector spent two years deciding whether AI belonged in ministry and has largely answered it. The live decision now, for pastors and operators alike, is governance of where: which tasks an agent may take on, and which a person has to own. Drawn well, that boundary reads less like hesitation about technology and more like a deliberate design choice, the kind that can be built into a system rather than left to good intentions.

Sources

• New Research: Pastors Are Using AI More Than You Think, Barna Group
• Majority of Pastors Use AI in Ministry Amid Concerns and Caution, Gloo
• AI Homilies? Never, These Clergy Members Say, The Advocate