A church chatbot that won't make things up

Ask Church can't invent doctrine: it answers only from your church's own published content, cites the source on every answer, re-checks each answer before sending, and says "I don't know" instead of guessing.

The real fear, named plainly

The thing keeping AI off most church websites isn't cost. It's the nightmare of a chatbot telling a visitor your church believes something it doesn't — or putting a sentence in the pastor's mouth he never said. That fear is correct. A general chatbot bolted onto a church site *will* do this, because it answers from everything it ever read.

How Ask Church makes that impossible by design

Four mechanisms, each independent of the last:

It only reads your content. Ask Church is grounded in your site, sermons, podcasts, and PDFs. It is not connected to the open internet for answers. There is no "general knowledge" it can fall back on to fill a gap.
It cites the source on every answer. The link is part of the answer. If there's no source, there's no answer.
It re-checks before it sends. A second pass verifies the answer against your own content before the person ever sees it. If the check fails, the answer doesn't go out.
It says "I don't know." When the answer isn't in your content, it tells the person and hands off to a real human. It does not improvise.

This is the difference between *asking* an AI to be careful and *building* it so it can't be careless. A prompt that says "only answer from church content" is a suggestion the model can ignore. A grounding gate in the code is not.

And before any of that — crisis safety

The one place a wrong answer is unforgivable is a person in real distress. Ask Church handles that first, in code: a message signaling acute distress — in English or Spanish — routes to the 988 Suicide & Crisis Lifeline, the Crisis Text Line, and your own pastoral-care team *before any AI runs at all.* The assistant never tries to be the counselor.

Frequently asked

Couldn't it still hallucinate?

The grounding gate is designed to stop exactly that: every answer is re-checked against your own content before it sends, and an answer with no source in your content isn't sent. When in doubt, it hands off to a person.

What if our website has a mistake on it?

Then the assistant will reflect your site — which is why you control the content. It answers from what you've published, and you can correct the source.

Does grounding make it slow or robotic?

No. It answers in seconds and in your church's voice. Grounding constrains where the answer comes from, not how it reads.