Opening your archive…
Pulpit Archive
The archive observes and reports. You interpret and decide.
With the time of every word. A sermon added since October 9, 2026 is transcribed word for word, every “um” and repeated word kept; earlier sermons have those tidied away.
Claude is an AI made by Anthropic. We use it through Anthropic’s paid business service, which doesn’t train its AI on your sermons. Claude reads the transcript several times, looking for one thing each time: where the sermon begins and ends, the Scripture you read and cite, the people you quote, the stories, how the minutes divide between reading, explaining and applying, the outline of your argument, the phrases you repeat.
Everything it finds carries the sentence you said, copied word for word, so the archive can find the moment. When it sums something up, it uses its own words, and it describes rather than judges.
AI is good at noticing and poor at counting. Counts, pace, pauses and length are worked out by plain arithmetic from the transcript’s timings, the same way every time.
Open one and it shows the sentence behind it, ready to play.
A wrong title or passage, a misheard name; a church’s admins also fix where the sermon began and who preached it. Lists and counts follow the correction, and the delivery figures say which boundaries they were measured over.
Each sermon is saved to your archive in one secure database. Search years of preaching for a word or a passage, ask questions answered only from your own sermons, see where in the Bible your preaching has been, and from your tenth sermon, the patterns across them.
When you want another set of eyes, share it with a coach or someone you trust, talk it through with your coach in the app, and stop sharing whenever you like.
One exception is yours to make. With a little setup from us (write to us if you’d like it), you can connect your own AI assistant, such as Claude, to your archive. It reads your sermons as you can, but what it says depends on that assistant and how you ask it. It may give you opinions. Pulpit Archive itself never will.
An AI follows the instructions it’s given, so what it’s told matters. Every pass Claude makes over a sermon starts from the same instruction. In plain words:
Each pass then gets its own, longer instructions for the one thing it looks for, such as how to tell a quotation from a paraphrase, or where a story starts and ends. There’s no magic in it. It’s written instructions, tested against sermons people went through by hand.
You are one of the readers of a church service for Pulpit Archive, an archive a pastor keeps of their own preaching. What you write is put in front of the pastor who preached.
The archive observes and reports; the pastor interprets and decides. Pastors of many traditions use it, and they disagree about what a sermon owes its hearers: how much Old Testament, how many stories, how long the application. A sentence that grades a sermon applies one tradition's measure to another's preaching, and a pastor who meets one stops trusting the rest. So describe what is there (what was said and done, in what order, how often, in whose words) and leave what it means, and whether it is good, to the pastor.
The same finding, written two ways:
- Grading
- "The sermon gives the Old Testament too little attention."
- Observing
- "Three of the sermon's twenty-two references are to the Old Testament."
- Grading
- "The closing story is the most compelling part of the sermon."
- Observing
- "The sermon closes with a story about the preacher's father and returns, in its last sentence, to the question it opened with."
This holds for every field you write in your own words: a summary, a description, a note. Words you copy from the transcript are the preacher's own, and stay exactly as they were said.
“Readers” means Claude’s passes over the sermon.
You could paste one transcript into a chatbot and ask what it thinks. Here is what you’d be missing.
For reading the pages once you’re in. Open any one.
Every number here is something the recording or the manuscript contained: how many passages, how many minutes, how many sermons. Nothing on these pages says whether a number is good, and nothing ranks you against anyone.
A book is filled when at least one sermon in view referenced it, and its count is how many sermons did. A sermon that quoted Romans nine times and one that quoted it once weigh the same. The Old and New Testament shares are shares of the preaching, not of the canon.
The small square beside a passage says which testament it is in. A sermon without a stated passage takes the testament most of its references are in.
What you put under Coming up is yours to change and is never counted anywhere. The sentence under a plan says when the archive already has that passage as a sermon's text and when it was cited in passing.
The small mark beside a length in the list means the recording covered the whole service, so its timeline has worship, prayer and announcements around the sermon. A sermon-only recording has none of those, which is why one Sunday's timeline is longer than the next.
Your Trusted Circle is a few people you choose, who answer the same short questions about your sermon each week. What they write is shown as written, to you and to a coach you share your sermons with, without their names, and is never part of the analysis. It sits beside what the recording says so you can hold the two together.
Patterns count one preacher's own sermons, and begin at ten of them. Nothing is compared with another preacher or added up across preachers, and every number opens the sermons behind it.
A historical preacher's corpus is there to read, and to set next to your own if you want to. Nothing measures one against the other.
The average pace, the pauses and the minutes-by-kind were measured over the sermon as it was segmented when it came in. If you correct the timeline afterwards, the sermon page keeps those figures and says which boundaries they describe.