Can AI write out a sermon as a faithful manuscript?
Yes, if it is written from the transcript and not from a memory of it, and if the person who preached can read it and recognise every point in the order they made it. A summary tells you what the sermon was about. A manuscript is the sermon.
Why a summary is not enough
Most tools that listen to a service hand back a summary: three paragraphs, the scripture references, the takeaways. It is useful for a newsletter and useless for the person who was not there. They do not want to know what the sermon was about; they want the sermon.
A spoken sermon written down word for word is not the answer either. Transcripts carry every repeat, every false start, the notices about the car park and the reminder about the offering. Nobody reads forty minutes of that. The thing a church actually wants sits between the two: the whole teaching, in the preacher's order, readable in one sitting.
What may go, and what must stay
The line between editing and rewriting is where a manuscript is either faithful or not, so it is worth stating what is on each side of it.
- May go: welcomes, notices, logistics, the offering appeal, the dismissal, the sentence started twice, the filler between thoughts.
- May be tightened: a point made three times in three ways can be made once, well.
- Must stay: every point, in the order it was made, under the preacher's own headings; the story the message was built on; the verses as quoted; names and places as said.
- Must never happen: a point the preacher did not make, a verse they did not quote, a smoother theology than the one they preached.
The fourth line is the one to test. A machine that improves a sermon has changed it, and a manuscript that reads better than the preacher sounds is a warning, not a compliment.
How a faithful manuscript is built
The reliable way is two passes over the transcript. The first reads the whole thing and finds the outline: where each section begins and ends, and what it is about. The second writes each section from the lines of transcript that belong to it, and nothing else. The headings come from the outline, so the structure is the preacher's and not the machine's.
Other languages should be translated from the finished manuscript, section by section, with the church's own glossary, rather than written again from each translated transcript. That way the German manuscript is built from the text written against the original speech, every language has the same outline, and a phrase the church has a settled translation for is translated that way.
A manuscript that is not finished should not be published. If the writing stops halfway, for whatever reason, the church should see a draft that says so rather than a document that ends mid-sentence.
How to check one in ten minutes
A very short talk produces no manuscript at all, and should not: a few hundred words of transcript cannot carry a structure, and a tool that writes one anyway is inventing.
How Sanctavo does it
Every translated service becomes, among its seven pieces, the full message: outlined from the transcript, written section by section under the preacher's own headings, with service notices removed and every point kept in its order. The other languages are translated from that manuscript with the church's glossary, so they share its outline. An unfinished manuscript is never published, a talk too short to carry one gets none, and everything waits for the church's review before it goes anywhere. Because the system uses AI, errors remain possible, and names, numbers and scripture references are the places to read first.