Signal

A machine can assemble a sermon now. It can find a passage, propose three movements, write a turn toward hope, and end on a sentence that sounds as though someone sat with it all week.

In June 2023, an experimental service at St. Paul's church in Fürth, Germany, gave the question a body. Four avatars led a full church through forty minutes of sermon, prayers, music, and blessing. The theologian who organized it, Jonas Simmerlein, estimated that about 98 percent of what the congregation saw and heard came from the machine. People laughed at some passages. Congregants were asked to rise; some spoke the prayers along, and others would not say the Lord's Prayer aloud. The service could be followed. Whether it could be joined was a decision each pew made alone.

Months earlier, at the close of 2022, Rabbi Joshua Franklin preached a sermon written by ChatGPT to his congregation in the Hamptons, then told them who had written it. The point of the experiment was larger than whether the words sounded convincing. A machine could sound convincingly like a rabbi; Franklin argued that it could not supply the empathy at the heart of his work.

Listen beneath both scenes and you hear the same unease. If the words arrive without the person having lived with them, what has been lost?

Pattern

John Piper tested the question on himself. He asked ChatGPT for an answer about the dangers of pastors using AI, written in his own theology and style. It returned 857 words in five seconds. He thought listeners might have accepted the result as his own. That was precisely what troubled him. Reading it as his own, he said, would have deceived them.

Carey Nieuwhof describes the same fault line from the working side. Preparing a sermon involves research, but also the slower work of wrestling with a text and with the people who will hear it. He sees real uses for AI in research and the development of ideas. His warning is that letting it do the preacher's work interrupts that formation, and that an undisclosed machine-written message violates the trust between preacher and congregation.

This is where the argument over tools becomes an argument over attention. Sermon preparation has always done two jobs at once. It produced the words, and it produced the person able to mean them. A pastor may use a model to locate a source, suggest a question, or expose a sentence that does not make sense; each answer still needs checking. But a model can also remove the hours in which a preacher discovers that a familiar passage is harder, stranger, or more demanding than expected. Those hours are often where the sermon changes its speaker before it reaches anyone else.

Implication

There is a difference between receiving help with the work and presenting work you did not do as the fruit of your own prayer and study. The congregation cannot see the preparation. It has to trust the person in front of it. When substantial parts of a message have been composed elsewhere, honesty about that help matters.

Scripture does not give us a rule for language models. It does ask that a pastor be able to teach (1 Timothy 3:2). That is a demanding human responsibility: to understand what one is saying, answer for it, and care for the people hearing it. A generated page can assist with some of the words. It cannot take that responsibility.

And the people in the pews can already get fluent religious language on demand. They can ask a model for an explanation of a verse, a prayer, or a word for a hard morning. A preacher's task cannot be to produce the same thing on a larger screen. What remains scarce is not the language; it is a person who has stayed with the text long enough to be answerable for it.

Posture

So the question before Sunday is more searching than Did you use AI?

What did it help you see? What did you check? What did you let it say for you? And have you stayed with the passage long enough to mean what you are about to say?

Those questions do not forbid the tool. They relocate the work to the hours no listener will ever see, which were never packaging around the sermon. They were the sermon.

Undersong

The machine can say the words. It cannot mean them.

Meaning is not a property of sentences; it is the mark the hours leave on the one who must stand behind them. Wherever an audience trusts that words passed through a person — the pulpit, the eulogy, the letter that matters — fluency has now become free, and the worth of speech moves to the one thing that cannot be generated: a person formed by having lived with what they say.

Core Pattern
Producing the words and being formed by producing them were always one job; a machine that takes over the first reveals that the second was the point.

What This Alters
The honest question about machine help shifts from "did you use it" to "what did you let it say for you," and disclosure becomes part of the work, not an apology for it.

Resonant Line
The machine can say the words. It cannot mean them.

Passages for Transmission

  • Sermon preparation has always done two jobs at once. It produced the words, and it produced the person able to mean them.
  • A model can remove the hours in which a preacher discovers that a familiar passage is harder, stranger, or more demanding than expected. Those hours are often where the sermon changes its speaker before it reaches anyone else.
  • The congregation cannot see the preparation. It has to trust the person in front of it.

Drafted from an author manuscript on machine-written sermons. The Fürth service, Rabbi Franklin's experiment, and the Piper and Nieuwhof positions were verified against the contemporary reporting listed in the references.