Lecture 11-1: The Place of AI in Sermon Preparation

2 Timothy 4:2 — "preach the word; be urgent in season and out of season; reprove, rebuke, and exhort, with all patience and teaching." (World English Bible)

Genesis 32:24 — "Jacob was left alone, and wrestled with a man there until the breaking of the day." (WEB)

1 Corinthians 2:4 — "My speech and my preaching were not in persuasive words of human wisdom, but in demonstration of the Spirit and of power," (WEB)

3.1 A tool that can do more than prior tools could

Every generation of preachers has had research tools appropriate to its era: a concordance to trace a Hebrew or Greek word across its occurrences, a set of commentaries representing a range of scholarly judgment, a Bible atlas or dictionary to fix a historical or geographical detail, and, more recently, searchable digital libraries that can surface a cross-reference or a patristic quotation in seconds that once took an afternoon in a library. No responsible theological tradition has treated the use of any of these tools as compromising the sermon's authenticity, because none of them supplies the preacher's conclusion. A concordance shows where a word occurs; it does not tell the preacher what to conclude the word means in this text, for this congregation, on this Sunday. The preacher still has to do the work of judgment that turns raw data into a preached word.

A generative AI system is a genuinely new kind of tool because it can, if asked, skip that step of judgment for the preacher and hand back a finished conclusion — an interpretation, an outline, an illustration, a full manuscript — dressed in exactly the fluent, confident prose a human writer would use, with no visible seam marking where the preacher's own thinking stopped and the machine's pattern-completion began. This is not a difference of degree from a concordance; it is a difference of kind, because it is capable of producing the entire visible product of sermon preparation, including the parts that were always supposed to be the site of the preacher's own encounter with the text.

3.2 What legitimate use looks like

Several categories of AI use in sermon preparation function as genuine extensions of long-permitted research aids, and this lecture treats them as legitimate without qualification:

Word studies and lexical questions. Asking an AI system to summarize the semantic range of a Greek or Hebrew word, or to list its occurrences in a given book, functions exactly like consulting a lexicon or a concordance — the tool surfaces information the preacher still has to weigh and apply.

Cross-reference and background lookup. Asking for parallel passages, historical background on a custom or place, or a survey of how a text has been read across church history functions like consulting a study Bible or a commentary series — again, information gathering, not conclusion-forming.

Structural brainstorming. Asking an AI system to suggest possible ways to organize a sermon's flow — three points, a narrative arc, a problem-response structure — can be a legitimate aid to the preacher's own organizational thinking, provided the preacher evaluates and substantially reshapes what comes back rather than adopting it wholesale, in the same way an editor's suggested outline is a starting point rather than a final product.

Clarity checking on a draft the preacher has already written. Asking an AI system to flag an unclear sentence, an overlong paragraph, or a place where the argument's logic seems to skip a step, applied to a manuscript the preacher has already substantively composed, functions like an editor's proofreading pass — it improves the expression of a conclusion the preacher has already reached, without supplying the conclusion.

3.3 What illegitimate use looks like

The same tool, asked a different kind of question, crosses into use this lecture argues against:

Generating the sermon's exegetical conclusions. Asking an AI system what a passage means, what its main point is, or what application it yields for the congregation, and then preaching that answer, substitutes the tool's pattern-completed synthesis of commentary-like text for the preacher's own encounter with the passage. The output may well be biblically accurate — commentaries themselves are a major part of what the underlying model was trained on, so a reasonable answer is likely — but its accuracy is not the point. The preacher has not exegeted the text; the preacher has retrieved someone else's exegesis, mediated through a system with no theological commitments, no pastoral relationship with the congregation, and no capacity to have actually understood what it produced.

Generating the sermon's content wholesale. Asking an AI system to write the sermon — an outline populated with the preacher's own developed content, illustrations chosen and shaped by the preacher's own pastoral judgment, applications aimed at the preacher's own knowledge of the congregation's actual struggles — and then preaching the result with minor edits, is the clearest instance of the line this lecture draws. What is preached is not, in any meaningful sense, the fruit of the preacher's own preparation, whatever the sermon's surface quality.

Outsourcing illustrations that claim personal or pastoral knowledge. A borderline case worth naming explicitly: asking an AI system to generate an illustration presented as though it came from the preacher's own experience, or as though it reflects specific pastoral knowledge of the congregation, compounds the authorship problem with a honesty problem (developed fully in lecture 11-3) — the congregation is being told something false about where the illustration came from, not only handed content the preacher did not originate.

The line is not tool use versus no tool use. It is: did the preacher's own judgment, wrestling, and conviction produce what is preached, using research tools (old or new) to inform that judgment — or did a tool produce the content, with the preacher's judgment reduced to selecting and lightly editing what came back?

3.4 "Preach the word" as a charge that resists outsourcing

Paul's charge to Timothy is characteristically compact: "preach the word; be urgent in season and out of season; reprove, rebuke, and exhort, with all patience and teaching" (2 Timothy 4:2, WEB). The verbs are active and personal — Timothy himself is to preach, to be urgent, to reprove, to exhort. Nothing in the charge is satisfied by Timothy having procured, from any source, a correct statement about the word; the charge is to a person, exercising personal, sustained, situationally responsive ministry ("in season and out of season," implying ongoing discernment of when and how to press a point, not a single delivered content-unit).

This matters for the present question because the charge cannot be satisfied by proxy even when the proxy's output is accurate. If an AI system produced a theologically flawless statement of what a passage teaches, and a preacher delivered that statement verbatim as the sermon, something would still be missing relative to what 2 Timothy 4:2 charges Timothy to do — not a missing fact, but a missing act of the preacher's own urgent, personal engagement with the text and the hearers. The charge is inherently first-personal in a way information transfer is not.

3.5 Jacob at the Jabbok: why this has to be a struggle, and why it has to be alone

Genesis 32:24 offers this lecture its central image, deliberately chosen over a more abstract appeal to "authenticity" or "originality." Jacob, on the eve of a reunion with the brother he had wronged and now feared, sends everyone and everything he has across the ford of the Jabbok, and Genesis reports starkly: "Jacob was left alone, and wrestled with a man there until the breaking of the day" (Genesis 32:24, WEB). The narrative does not explain in advance who the assailant is; it simply insists on two facts that matter for this lecture — Jacob is alone, and the encounter is a wrestling match that lasts the whole night, ending only at daybreak with Jacob wounded and renamed.

Applied to sermon preparation, the image is not a sentimental claim that preaching should feel hard for its own sake. It is a claim about what the encounter with a text is supposed to be: something the preacher undergoes personally, without a delegate to stand in, that can leave the preacher marked by the encounter in a way that shapes how the resulting word is delivered. A preacher who has genuinely wrestled with a difficult text — sat with its ambiguities, resisted an easy resolution, prayed through a passage that indicts the preacher's own life before it indicts anyone else's — preaches differently than a preacher presenting a synthesis someone else produced, however accurate that synthesis is. The wrestling is not merely instrumental to producing correct content; it is part of what forms the preacher into someone able to preach that content with the weight it should carry. A generative AI tool cannot wrestle at the Jabbok on the preacher's behalf, not because it lacks sophistication, but because the struggle itself is not a step that can be completed by anyone but the one who is to preach the result.

3.6 A standard AI-assisted eloquence cannot meet

Paul's account of his own preaching in Corinth supplies a further, sharper test. "My speech and my preaching were not in persuasive words of human wisdom, but in demonstration of the Spirit and of power" (1 Corinthians 2:4, WEB). Paul is explicit that he could have preached with more polished rhetoric — he had access to the conventions of Greco-Roman oratory and elsewhere shows he could use them — and deliberately did not lean on them, so that the Corinthians' faith would rest "not in the wisdom of men, but in the power of God" (1 Corinthians 2:5).

This is not, this lecture argues, a general suspicion of eloquence or good writing; Paul is a considerable rhetorician in his letters. It is a specific claim about where the persuasive power of true preaching actually comes from, and a warning against confusing polished, persuasive language with "demonstration of the Spirit and of power." An AI-assisted sermon can achieve extraordinary fluency and even genuine rhetorical polish — the underlying systems are, among other things, very capable stylists. But fluency is precisely the category 1 Corinthians 2:4 sets aside as insufficient, even when it names the very thing an AI system is best at producing. A generatively eloquent sermon can fail Paul's test not because it is poorly written, but because eloquence was never the test to begin with, and a tool optimized for eloquent-sounding output cannot, by the nature of what it is, produce "demonstration of the Spirit and of power" as a byproduct of its fluency. That demonstration, on Paul's own account, came from somewhere else in his own preaching — his weakness, fear, and trembling (1 Corinthians 2:3) — not from any measurable property of the sentences themselves.