How our studies are made
What a Deep Study is
A Deep Study is a short set of notes attached to a verse. It has five parts: the verse in plain words; the historical setting; the key words in Hebrew or Greek with what they mean; a handful of related verses; and a thought for today. It is meant to take two or three minutes to read. Thousands of verses have one at launch, across all 66 books of the Protestant canon, and more arrive with every update. Not every verse has one yet.
Where the material comes from
Every study is built from works whose copyright has expired or that are openly licensed:
- Matthew Henry's Concise Commentary (public domain);
- Jamieson, Fausset and Brown's Commentary (1871, public domain);
- Strong's Exhaustive Concordance and its Hebrew and Greek dictionaries (1890, public domain);
- cross references from OpenBible.info (CC BY 4.0), which we filtered and ranked.
Historical and linguistic claims in a study are supposed to trace back to these sources. The study is written against the verse text and quotes the translations word for word.
How a study is written
We do not have a team of writers, and we say so. Studies are drafted with large language models (Anthropic's Claude), working from the sources above under a fixed style guide. The style guide sets the structure, the length, the warm and plain tone, and a list of things a study must not do: no invented facts, no promises of outcomes, no shaming, no taking sides where Christian traditions honestly differ, and no "as an AI" voice.
All of this happens in advance, on our side, before the app is built. Studies are packaged into the app and shipped with it. Nothing is generated while you read. The app never sends your prompts, your reading history or any personal data to an AI system, because there is no AI system in the app to send it to. Everyone sees the same study for the same verse, offline.
How a study is checked
Each draft passes three layers of review before it ships.
- Automated checks. Software verifies that every verse reference exists and is written correctly, that no related verse points back to the verse itself, that every Strong's number quoted is one that actually occurs in the verse, that every quotation from a commentary is found word for word in the source, and that the text contains no links, no formatting leftovers and no AI self-reference. A draft that fails any check is rejected and rewritten.
- An independent review panel. Each study is then read by a panel of six separate AI reviewers, each with one job: a fact-checker who compares every claim against the sources; three readers who look for anything a Catholic, an Evangelical or an Orthodox Christian would find one-sided or unfair; a pastoral reader who watches tone (no guilt-tripping, no promised outcomes); and an editor who checks language and clarity. Each reviewer gives a score and a written reason. An arbiter collects the verdicts; studies that fail go back for regeneration with the panel's notes attached, and are reviewed again. In the current corpus about 2 percent of studies failed the first round and were redone.
- Spot checks by the maker. The maker of Sacred Feed reads a sample of studies from each batch, with a bias toward the verses where reviewers disagreed, and sends anything doubtful back.
We want to be exact about what this is and is not. The reviewers are AI models, not clergy or scholars. The human in the loop is the maker, who is not a theologian. We do not claim the studies are "scholar-reviewed" or "approved" by anyone. We claim that they are sourced, checked against those sources several times, and corrected when you tell us something is wrong.
Where traditions differ
Christians read some passages differently, and some have different books in their Bibles. The style guide asks a study to describe the main readings without choosing one and to say "traditions differ" where that is the case. The Catholic, Evangelical and Orthodox reviewers exist to catch the places where a draft quietly adopted one tradition's framing. They do not catch everything. If a study reads as one-sided to you, that is exactly the kind of report we want.
Limits
- Studies may contain mistakes: a wrong cross reference, a loose paraphrase, a historical detail the sources do not support.
- They lean on commentaries written in 1706 and 1871. Those are rich but dated, and they are Protestant works. We try to balance that through the review panel, not by hiding it.
- They are educational notes, not Scripture and not the official teaching of any church, and they are not pastoral or medical advice. See the Terms of Use.
Reporting a mistake
Every study has a Report an issue option at the bottom. Tell us what is wrong in a sentence; reports go to app@sacredfeed.com. We check the claim against the sources, correct or withdraw the study, and ship the fix in the next app update. You will not hear a marketing reply; you will hear what we found.
Privacy
Because studies are prepared in advance and shipped inside the app, no personal data about you is ever sent to an AI system, ours or anyone else's. Details in the Privacy Policy. Sources and licenses are on the Licenses page.