Lecture 15-2: Drafting a Church AI Use Policy — A Practical Exercise

1 Corinthians 14:40 — "Let all things be done decently and in order." (World English Bible)

Luke 14:28-30 — "For which of you, desiring to build a tower, doesn't first sit down and count the cost, to see if he has enough to complete it? Or perhaps, when he has laid a foundation, and is not able to finish, everyone who sees begins to mock him, saying, 'This man began to build, and wasn't able to finish.'" (WEB)

Proverbs 15:22 — "Where there is no counsel, plans fail; but in a multitude of counselors they are established." (WEB)

3.1 Why a policy, and why now, not after something goes wrong

Most congregations that use AI tools at all are using them the way most institutions first use any new technology: informally, tool by tool, decision by decision, with no single document anyone could point to that says what is and is not acceptable. This is not a moral failure so much as an ordinary pattern — policy usually follows practice, and practice usually runs ahead of anyone noticing a policy is needed. But 1 Corinthians 14:40's "decently and in order" is not a call issued only after disorder has already caused harm; Paul writes it into a passage governing worship practices that were, at the time of writing, still working reasonably well but showing early signs of the disorder that unregulated enthusiasm produces. The pastoral point generalizes directly: a policy adopted before a data exposure or a pulpit controversy is governance; a policy adopted only after one is triage. Chapter 15's opening claim — that the most consequential decisions are quiet and cumulative, not dramatic — is itself the argument for writing the policy now, while the stakes of any single decision still feel small enough to think about clearly.

3.2 Permitted uses

A workable policy names categories of task, not brand names of software, because tools change constantly and a policy tied to a specific product is obsolete within a year. The relevant category, drawing on the boundary this course has used since chapter 11, is whether the task is properly informational, logistical, or preparatory — in which case AI assistance is ordinarily appropriate — or properly relational, pastoral, or sacramental — in which case it is not, regardless of how good the tool has become. Concretely, permitted uses typically include: scheduling and calendar coordination; drafting routine, non-doctrinal communications (a facilities announcement, a volunteer sign-up reminder) subject to a human review before sending; first-pass transcription of a meeting or sermon for archival or captioning purposes; and research assistance of the kind lecture 11-1 already commended for sermon preparation — word studies, cross-reference lookup, structural brainstorming — provided the resulting content and conviction remain the preparer's own.

3.3 Prohibited uses

The prohibited list should be drawn with equal specificity, and for the same reason: a vague prohibition ("don't misuse AI") gives no one anything to actually check their own practice against. Drawing on the course's argument across chapters 11 through 14, prohibited uses should include, at minimum: using AI output as a substitute for pastoral counseling in any situation involving crisis, abuse disclosure, or acute distress (chapter 12's argument in full); generating sermon content or exegetical conclusions wholesale rather than using AI as a research aid (11-1's authorship line); making any decision about a specific member — discipline, pastoral follow-up priority, membership status — based solely on an algorithmic flag without human judgment and relationship-based knowledge of the person (extending 15-1's caution about risk-flagging); and using AI to generate doctrinal or catechetical content presented to the congregation without a qualified person's review, since 14-2's warning about confident, fluent, and sometimes wrong AI output ("hallucination") applies to institutional teaching with special force.

3.4 Disclosure requirements

Disclosure is the hardest section to write well because it resists a single blanket rule, as 11-3 already argued in the specific case of preaching: research assistance ordinarily needs no more disclosure than a commentary consulted in the study, while content substantially AI-drafted and presented as the preacher's own reflection would need disclosure or should simply not happen. The same logic extends across ministry areas, and a workable policy states the standard, not just an example: disclosure is owed whenever a person's trust in what they are receiving depends on assumptions about its origin that AI involvement would unsettle if known. A member reading a pastoral letter assumes it reflects the pastor's own attention to their situation; if AI drafted the bulk of it from a template, disclosure — or simply declining to use AI for that specific letter — protects a trust the member has already extended, on the standard 11-3 already set with Ephesians 4:25's call to truthfulness between neighbors.

3.5 Data-handling standards

This section operationalizes lecture 15-1's argument rather than repeating it. At minimum it should require: an inventory of which church systems use AI-driven features and what vendor processes the underlying data; a read of the vendor's actual data-retention and model-training terms before adoption, not after a complaint; a default of not routing the most sensitive categories of pastoral data (counseling intake, abuse disclosures, financial hardship details) through any AI-enabled system without an explicit, documented exception approved by a named leader; and a review cycle — annual is reasonable — since vendor terms change and a system approved two years ago may have quietly changed what it does with submitted data since.

3.6 A note on process, not just content

Proverbs 15:22's "in a multitude of counselors they are established" bears directly on how this policy should be produced, not only on what it should say. A policy drafted by one staff member and circulated for signature is not the same exercise as one built with input from those who will actually administer it (the office staff who will decide, in practice, whether a given form gets routed through an AI-summarizing intake system) and those most exposed by it getting wrong (a pastoral care team handling the most sensitive disclosures). The counting-the-cost discipline of Luke 14:28 is easiest to skip precisely when only one person is doing the counting.