AI & Your Data
Some features in Shepherd's Hook use AI to save your leaders time — summarizing a person's history, drafting a note, or reading a receipt. Because these features touch your church's information, we want to be completely clear about how they work and the limits around them. This page is the full picture.
1. AI is optional and off until you turn it on
No AI feature runs until someone at your church opts in. Consent is recorded, and every AI feature re-checks that consent each time before it does anything. If consent hasn't been given, the feature simply doesn't run. You can withdraw consent, and the features stop.
2. One provider: Anthropic (Claude)
All AI processing in Shepherd's Hook is done by Anthropic, the maker of Claude. We don't send your data to a patchwork of AI companies — it's this one provider, under Anthropic's commercial terms. (Shepherd's Hook is built by a team that works closely with Claude, which is part of why we chose a single, accountable provider.)
3. What actually gets sent — and what doesn't
When you use an AI feature, only the specific information that feature needs for that one task is sent to be processed, and only at the moment you use it. Nothing is sent in the background, and your whole database is never handed over. For example:
- Person summary: the notes and contact history for that one person, to write a short narrative.
- Draft / recap help: the specific messages or text you're asking it to summarize or draft.
- Receipt reading (Expenses): the image of the receipt you uploaded, to pull out the amount and vendor.
- Check-in setup: only a blank form's field labels — not anyone's submitted answers.
4. Your data is not used to train AI models
The information sent for these features is not used to train Anthropic's AI models. It's used to do the task you asked for, and then it's not retained to improve the model.
Being precise for your attorney: this no-training protection is a contractual commitment that applies to data sent under Anthropic's commercial/business terms — it is the legal agreement with the provider, not a switch in Shepherd's Hook's own code. We rely on that agreement, and we don't send your data to any consumer AI product that would carry weaker terms.
5. Guardrails
- Consent gate on every feature, re-checked each time.
- Usage limits to prevent runaway or abusive use.
- Minimization — each feature sends only what it needs, not everything.
- Human judgment stays with your leaders. AI drafts and summarizes; it does not decide who gets removed from a list, approve money, or take pastoral action on its own. A person reviews and acts.
6. A note on accuracy
AI output can be wrong or incomplete. Treat summaries and drafts as a helpful starting point to review, not as a final record. The authoritative information is always what your leaders enter and confirm in the app.
7. Questions
Email support@shepherdshook.org. See also Sub-processors and the Privacy Policy.
Shepherd's Hook