How we think about this tool

What we believe about this tool

Sermonwright is a study partner, not a sermon writer. We built it on three theological convictions, not just product policies:

1. Biblical text interprets itself. Exegesis starts with what the passage actually says — not with commentary traditions or historical speculation about what a word might have meant in a different century.
2. Historical references support. When we cite historical context, authorship, dates, original-language meanings, or traditions, we label them for verification. Real scholarship is a gift to preaching; unchecked claims are a threat to it.
3. Non-additive. We will not put words in the passage that aren’t there. We will not invent cross-references, statistics, historical events, or quotes from church fathers. When the model is uncertain, it says so.

How the validator works

Every generation is passed through a second model whose only job is to flag claims. It reads the passage you provided as ground truth, then catalogs every factual claim in the output and classifies each:

• Clean — grounded in the passage you provided, or clearly framed as hypothetical.
• Verify — plausible, but confirm before preaching. Historical claims, Greek or Hebrew word studies, cross-references to other passages, and views attributed to traditions all land here.
• Flag — likely fabricated. Altered quotes, invented attributions, made-up statistics, or cross-references where the cited verse doesn’t contain what’s claimed.

Scripture quotes specifically get a second, deterministic check — a simple string comparison against the passage you pasted. If the model paraphrases something as a direct quote but the words don’t match, we catch it.

What we won’t do

We will not invent quotes. We will not invent statistics or “studies show” claims. We will not invent historical events or manuscript discoveries. We will not attribute views to named scholars or church fathers unless you have provided the source. When the model is tempted to reach beyond what the text says, the discipline is: name the uncertainty, or say nothing.

When the model slips — and it sometimes does — the validator catches it and flags it for you.

What you still do

Exegesis from AI is a draft, not a sermon. You still own the theology, the ministerial judgment, the life-in-the-room behind the words. Our job is to give you a strong starting point and flag the places where your own work is still required. Your job is everything else.