Hi, this is Pastor Chris. I want to tell you how these materials get made, because “made with AI” now covers everything from a machine writing a document by itself to somebody using spell‑check, and those are not the same thing.
I could not build any of this on my own — the website, the HTML, the layouts, the printed guides. I learned HTML back in the mid‑90s! I have absolutely no business making these kinds of things. All of that side of it is the model.
Over the past year I built what is called a RAG — retrieval‑augmented generation. I went back through my old rhetoric, homiletics and hermeneutics categories, turned them into a fixed vocabulary, and then had AI read my sermons, other preachers’ sermons, and a good deal of other material — old commonplace books and the like — and break all of it down into a database the model can actually search.
Not just full texts. Every sermon gets split into functional units and tagged by what each one is doing — exposition, illustration, application, pastoral aside — along with its citations, its quotations, and the moves it makes across the canon. It runs to about fifteen hundred sermons now.
So a lot of what I do these days is prompting the model to go back into that database and find material I only sort of vaguely remember. If you want to see the vocabulary the whole thing is built on, it is public: theshepherdsguild.com/glossary.
If you are reading something like this study guide, the chances are good that I did not write these sentences. I edited them, and I certainly prompted them with specific instructions. I work at the structural level: I think about what needs to be said here or there, what order it goes in, what is missing, what is wrong. Then I read what comes back and send it out again.
The best metaphor I know for it is that I am something like an editor on a newspaper floor. I make the assignments, I look over the copy, I shape the stories, and I kill the ones that are not working. I am not the one at the keyboard writing the paragraphs. But nothing runs that I did not send back at least once.
Those are me, and I think you could tell by listening to them. I have been doing this long enough to have my own process. I will use AI for research now and then — usually leaning on that same database — and as a kind of interlocutor while I am working out delivery. But the sermon itself is mine.
This particular guide took about eighteen hours of my direct work across two days — roughly twelve on the first and six on the second. Most of that was reading, disagreeing, and rejecting drafts. I mention it because there is an assumption that a tool like this makes the work disappear. It does not. It changes what the hours are spent on.
If you have concerns about this process, I would love for you to read an entire transcript of the actual conversations that went back and forth over the development of this project. After reading it, I would be happy to discuss whether you feel this was a legitimate use of the tool. Please be prepared to set aside significant time — it runs to about thirty‑four thousand words.