Our AI Principles

AI touches enough of church.software — and enough of what a church is entrusted with — that we didn't think it was right to bury it in a footnote of our Terms of Service. This page is a separate, direct conversation about how we build with AI, what safeguards surround it, and how we hold ourselves accountable when it falls short.

1. We Understand Your Concerns

Pastors and church administrators have good reason to be cautious about AI, and we don't think that caution should have to be argued away. The concerns we hear most often — and take most seriously — are:

• ✓Reliability. AI-generated responses can be wrong, out of context, or confidently stated when they shouldn't be.
• ✓Privacy. Churches hold some of the most sensitive information that exists about a person — prayer requests, pastoral care notes, financial giving, family circumstances — and sending any of that to a third-party AI system is not a decision to take lightly.
• ✓Over-reliance. There's a real risk in letting a system initiate or complete tasks on a church's behalf without a person in the loop — not because the software might fail, but because ministry decisions deserve human discernment.

These aren't hypothetical objections to us — they're the starting constraints we designed around, not concerns we addressed after the fact.

2. The Power — and the Limits — of Generative AI

We also believe it would be a disservice to pretend generative AI isn't genuinely useful. Done well, it can surface a more thorough first draft than a volunteer squeezing in sermon prep between a day job and a family — a service plan built from real play history and a sermon's theme, a first-pass email newsletter, a pastoral call distilled into a clear follow-up task. That's real capability, and we build toward it.

But capability is not the same as authority. Every AI feature in church.software is designed around one rule: AI output is always a suggestion to be reviewed, never an action taken unchecked. A suggested setlist can be swapped. A drafted email is reviewed before it sends. A call summary is a starting point for a human follow-up, not the follow-up itself. If a feature can't be built that way, we don't ship it that way.

3. What Guardrails Are, and Why They Matter

"Guardrails" is a term we use often, and we think it deserves a plain explanation rather than a buzzword. In church.software, guardrails are the specific, concrete checks placed around every AI interaction — before a request leaves our systems and before a response ever reaches you:

• ✓De-identification before send. Personal identifiers are stripped out of a request before it ever reaches an AI model, so the model is working with the minimum information it needs, not a person's full record.
• ✓Output screening. Responses are checked against our content and ministry guidelines before they're shown to you — and anything a model invents that isn't backed by your real data (a song that isn't in your library, a fact that isn't in your records) is caught and discarded, not displayed.
• ✓Narrow scope. Each AI feature is built to do one bounded, auditable job — suggest a setlist, draft an email, summarize a call — rather than given broad, open-ended authority to act across your church's data.
• ✓A safe fallback. If an AI step fails, times out, or doesn't clear a check, the feature falls back to a plain, deterministic result instead of guessing — AI is always an enhancement, never a hard dependency.

Guardrails exist for one reason: so a ministry leader can trust that whatever AI touches has been bounded, checked, and is attributable to a real, inspectable process — not a black box making decisions on your church's behalf.

4. Safe. Responsible. Ethical. Biblical.

We hold every AI feature to four standards, and we mean each word specifically:

• ✓Safe — protected by the technical guardrails above: de-identified inputs, screened outputs, and encryption at every step in between.
• ✓Responsible — transparent about what a feature does and doesn't do, and accountable when it gets something wrong, rather than hiding behind "the AI did it."
• ✓Ethical — fair, free of manipulation, and never presented as a human when it isn't one.
• ✓Biblical — held to a standard the other three don't cover on their own.

We add "biblical" deliberately, and as a distinct fourth standard rather than a synonym for "ethical." A tool can be technically safe, procedurally responsible, and broadly ethical, and still be used in a way that is spiritually careless. So even for tasks that look entirely administrative — drafting a newsletter, suggesting a song, summarizing a phone call — we ask a second question beyond "did this work correctly?" We ask whether the way it was done reflects a sanctified use of the tool: set apart for God's purposes, not merely efficient. That theological check is applied across the platform, not reserved for features that look overtly "spiritual" on the surface — because in a church, we don't believe any task truly is only administrative.

5. We Are Not Perfect — We Need You

Every safeguard on this page is something we've built in good faith, and none of them makes us infallible. AI can still misfire, misunderstand context, or be misused in a way we haven't anticipated. We would rather hear about it from you than have it go unreported.

If you ever encounter a wrong or concerning AI output, have even the slightest unease about how AI is being used in your church's data, or need to report misuse of any kind — please reach out. A real person reads and responds to every message that comes to us about AI; you will not get a bot in return.