Philosophy
How We Think About AI
Sermon preparation is a spiritual discipline. It involves prayer, study, wrestling with the text, and the leading of the Holy Spirit. That process belongs to a pastor. No machine should try to replace it.
Sermon Analyst exists to do what a human can't. You or a coach can review a few sermons or recall major themes. A machine can do this on a much larger scale to find patterns across years of preaching you can't easily see on your own.
Large Language Models (LLMs, a type of AI) are what make this possible, but our design philosophy is rooted in what we think they are actually good at — and what they aren’t.
What AI does well
AI is fundamentally a pattern-recognition engine. It’s good at reading large volumes of text and identifying structure: recurring themes, gaps in coverage, how often certain topics or passages appear, whether a sermon’s introduction and conclusion connect. It can compare a draft against a body of past teaching and surface observations.
These are the same kinds of things a thorough preaching coach or a diligent seminary professor might notice if they had the time to review every sermon you’ve ever preached. AI just does it much faster.
Sermon Analyst uses AI in this way: to aggregate, synthesize, and find patterns in your preaching. Everything the system produces is grounded in what your church has actually taught — your sermons, your stated goals, your context.
What AI does not do well
AI cannot discern spiritual truth. It cannot be convicted by a passage of Scripture. It cannot sense that a congregation needs to hear something difficult, or that a sermon needs to linger on a moment of grace rather than move to the next point. It cannot pray.
These aren’t limitations that may be solved with better models. They’re category differences. The work of the Holy Spirit in sermon preparation is not an information-processing problem, and treating it as one misunderstands both the technology and the theology.
This is why we don’t build features that write sermons or assess theological correctness. We think this would be a mistake — not because the output is always bad, but because AI cannot and should not replace a Spirit-led person in shepherding a congregation.
Not a chatbot
One thing that distinguishes Sermon Analyst from a general-purpose chatbot is that every input is explicit. The system analyzes your sermon library, your own provided context or instructions, and whatever you choose as a prompt. That’s it.
It doesn’t learn from your past queries. It doesn’t adapt to your behavior over time. It doesn’t build a hidden memory of your preferences and subtly shift its analysis to match. Every time you run a report, the system approaches your content fresh, using the same consistent foundation.
This is a deliberate design choice. A tool that gradually adapts to your biases can end up reinforcing them. By keeping the inputs consistent, we make the system more predictable and objective.