Chapter 15: Practice Parameters

Chapter fourteen named what must stay human in formation and care. This chapter consolidates operating defaults—not eternal law. Models will change; your board might not. Good parameters are written in pencil, reviewed quarterly.

70 / 30 (governance heuristic, not science)

Treat 70/30 as a fence, not a finding: roughly more human judgment than unreviewed machine text on publish-tier material. I am not leaning on any single survey as scripture; for time-on-task, full-model drafting, and self-reported outcomes in one marketing-style blogging study (Orbit Media, 2025), see chapter six—where that evidence first appears in this book. For how the same discipline maps to a repeatable publish chain, see chapter sixteen. The point here is posture, not a statistic.

Public language: AI expands and structures; humans judge, correct, and take responsibility.

Agent roles: assistants, archivists, translators—not authors

Assistants propose outlines, counterarguments, edits—you own acceptance.

Archivists index, tag, retrieve—truth still lives in curated sources, not embeddings alone.

Translators adapt format (talk → article) while you re-assert thesis sentences in your voice.

Authors of moral and theological weight remain human. COPE-style norms are explicit: AI tools are not authors; people are accountable for what ships (COPE guidance).

For movement leaders in public view, approval of AI-affected voice still belongs with the leader and whatever small circle they trust—tools do not inherit your reputation.

A workable spectrum (compressed)

Instead of eight micro-levels, think three bands:

1. Human-led, machine-assisted — research, editing, structure; you are visibly the source.
2. Human-directed, machine-generated — model drafts from your brief; heavy redraft; disclosure.
3. Machine-led — red zone for mission-critical publication defaults; at most R&D sandbox, never silent ship.

Different domains ride different bands: admin comms may tolerate band 2; liturgy, frontline counseling, and other care-shaped practices stay band 1 or human-only—not because models are “wicked,” but because the goods at stake are practice-shaped in ways I named in chapter fourteen (prayer, presence, slow moral time).

One organizational parameter I keep seeing omitted: during structured experimentation (the phase I unpack in chapter seven), some teams adopt an explicit moratorium on publishing AI-affected or AI-generated content—internal drafts and documented learnings only—until governance and reviewer roles exist. Panic after a public mistake is expensive; a written publish rule is cheaper.

Movement leaders: content flywheel (example only)

A compressed example only—the full capture → verify → publish chain lives in chapter sixteen:

Talk → transcript → human thesis → machine-assisted shape → human rewrite → named reviewers → publish → network amplification (mutual links, shared courses where integrity allows). The principle: workflow > sparkle.

Churches: four rooms on one campus

• Worship: lyrics projection OK; prayer voice stays human.
• Discipleship: AI-generated discussion prompts reviewed by a teacher who knows the group.
• Comms: templated newsletters checked for tone-deaf fluency.
• Admin: scheduling bots—celebrate them.

Nonprofits: triage table

Parameters expire

When vendors ship new models, re-run a red-team afternoon: hallucination checks, citation checks, voice drift checks. Parameters are a standing discipline, not wallpaper—they should advance with your maturity stage (again, chapter seven), not float as a one-off workshop poster.

Next: apply the same honesty to content pipelines (chapter sixteen) and the harder soil of formation (chapter seventeen).

---

Reflection questions

1. Map your top five public artifacts to bands 1–3. Any surprises?

2. Who is the named “human author” for each external channel?

3. Do you have a written rule for what may be published during experimentation—and who may override it?

4. When is your next quarterly parameter review—and what changed since last quarter?