Pope Leo XIV Warns Against AI Homilies as U.S. Pastor Use Hits 24%

Pope Leo XIV’s February 19, 2026 warning against preparing homilies with artificial intelligence remains the Vatican’s clearest public intervention in this debate. In the official record of his meeting with Rome’s clergy, he urged priests to resist the temptation, arguing that unused intelligence weakens and that AI cannot share a preacher’s faith.

The significant update is that adoption outside the Catholic priesthood has continued despite similar concerns about authenticity. Barna’s June 2026 findings report that 24% of surveyed U.S. Protestant pastors use AI to write or edit sermons, based on an online survey of 442 pastors conducted in December 2025.

The papal warning was about authorship, not every digital tool

Leo’s remarks were narrower than a blanket rejection of technology. He acknowledged that AI and the internet were already present in priestly life, but drew the boundary at preparing the homily—the part of ministry where a priest interprets scripture for a particular congregation and speaks from lived faith.

His reasoning had two distinct parts. First, delegating composition can erode the intellectual work expected of the preacher. Second, even a fluent and doctrinally plausible text cannot supply the personal religious experience that a homily is meant to communicate. His central formulation was categorical: “AI will never be able to share the faith.”

The setting matters. Leo was answering a question about serving communities whose circumstances continually change. Immediately before discussing AI, he stressed that a priest must know the particular place and people he serves rather than transplanting a formula that worked elsewhere. The objection therefore reaches beyond factual accuracy: a generated sermon may sound polished while remaining detached from the congregation that will hear it.

The published address is an exhortation to Rome’s priests, not a technical policy document or a universal enforcement regime. It does not define prohibited software, prescribe penalties or explain whether limited uses such as translation, transcription or research are acceptable. Describing it as a comprehensive Vatican ban would overstate what the official text says.

The newer data show a divided market, not wholesale automation

The 24% figure does not mean that a quarter of pastors hand complete sermons to a chatbot and read the result unchanged. The survey question combines writing and editing, while the accompanying analysis describes AI primarily as a preparation partner used to surface commentary, explore theological angles or sketch an outline.

That distinction is commercially important for companies selling generative tools to religious organizations. The more defensible market is not autonomous preaching but supporting work around it: brainstorming, biblical research, discussion questions, graphics and administration. In the same survey, 50% of pastors reported using AI for brainstorming, 36% for biblical or theological research, and 34% for small-group questions or administrative tasks.

The headline comparison with 2024 also requires care. The earlier 12% measured pastors who said they were comfortable using AI to write sermons; the newer 24% measured pastors who said they actually write or edit sermons with it. Those are different questions, so the figures suggest movement but do not establish a precisely measured doubling of identical behavior.

Resistance has not disappeared alongside adoption. In the same research, 71% of pastors described themselves as cautious about AI, 40% as conflicted and 40% as skeptical. The emerging market is therefore defined by simultaneous use and distrust: clergy may adopt tools for efficiency while remaining unwilling to treat generated language as spiritual authority.

A real AI service exposed the difference between novelty and presence

The consequences become clearer when generated material moves from a private study into public worship. In March 2025, a Lutheran church in Helsinki presented a 45-minute experimental service assembled with several AI systems. The Associated Press account of the Helsinki experiment says AI wrote sermons and some songs, composed music and produced video avatars, while live participants still sang hymns with an organ.

The organizers imposed explicit boundaries: AI did not perform the Eucharist or pronounce absolution, and clergy said outputs required human editing and fact-checking. Attendees found the event interesting, but some described it as distant, impersonal or closer to a performance than a service. That response illustrates the operational risk behind Leo’s theological argument: successful content generation does not guarantee that listeners experience a speaker as addressing them.

The experiment also shows why a simple “AI or no AI” framing is inadequate. The same clergy who rejected automated sacramental acts saw possible value in bookkeeping, locating relevant verses and researching speeches. Different tasks carry different expectations of accountability, intimacy and personal witness.

The practical boundary is responsibility for the message

For a church deciding how to govern these tools, the central question is not whether AI touched a document. It is whether the minister can take informed responsibility for every claim, interpretation and pastoral implication delivered to the congregation.

A workable internal distinction can separate assistance from substitution:

• Lower-risk assistance includes transcription, formatting, administrative drafting and generating search terms that lead back to sources a person verifies.
• Review-intensive assistance includes outlines, translations, summaries and theological research, where omissions or invented references can alter meaning.
• High-risk substitution occurs when generated prose replaces the preacher’s own interpretation, knowledge of the congregation or response to a specific pastoral situation.

This framework does not resolve every theological disagreement, but it makes accountability visible. A minister should know which passages were generated, verify citations and scriptural context, remove confidential information from prompts, and be able to explain the final argument without relying on the model that produced it. Disclosure may also be appropriate when AI materially shaped what a congregation is being asked to receive as personal testimony or pastoral judgment.

The unresolved conflict is becoming more precise

Leo’s intervention has not stopped clergy from experimenting with AI, nor does the evidence show that pastors have broadly surrendered sermon authorship. Instead, the boundary is shifting toward a task-by-task division: machines can accelerate preparation, but many religious leaders reserve interpretation, spiritual testimony and responsibility for a human speaker.

That division will be difficult for vendors to encode because the same feature can serve different purposes. An outline can prompt a minister’s own thinking or become a nearly finished sermon with a few edits. The decisive factor is the workflow around the output—who verifies it, who adapts it to the community and who remains answerable for what is said from the pulpit.

Also read:

• Did the Pope’s Anti-AI Encyclical Get (Partially) Written by AI? The Pangram Detector Says 46%
• Musk vs. Altman: Silicon Valley’s Most Explosive Showdown Finally Hits the Courtroom