- Home
- How answers are built and checked
Methodology
How answers are built and checked
Question interpretation, tradition selection, source retrieval, citation verification, neutrality safeguards and the limits of AI.
This page explains, in plain terms, how The Scripture Guide turns a question into nine sourced perspectives, what safeguards are in place, and where the method has limits. It is the long version of How It Works.
1. How questions are interpreted
A question is read as an invitation to describe, not to decide. “What happens after death?” is treated as “What does each tradition teach about what happens after death?” Leading or loaded phrasing is answered neutrally. Questions outside the scope of religious teaching (for example medical or legal advice) are redirected to the traditions' own teachings on the subject rather than answered as advice.
2. How the nine traditions were selected
The Guide covers Christianity, Islam, Judaism, Buddhism, Hinduism, Sikhism, the Bahá'í Faith, Taoism and Shinto: a set that spans the Abrahamic, Indian and East Asian families and includes traditions with a single canonical scripture as well as traditions (Hinduism, Shinto) where authority is spread across many texts and practices. Inclusion is not an endorsement, and the order in which traditions appear is fixed and carries no meaning. Each tradition is described in its own vocabulary, and major internal divisions — Catholic, Orthodox and Protestant; Sunni and Shi'a; Orthodox, Conservative and Reform; Theravada and Mahayana — are named when a question turns on them.
3. How sources are retrieved
There are two kinds of content on this site. Editorial pages (the question library, tradition pages and Scripture Explorer) are written by people, and every passage was chosen and checked by hand. Live answers to your own questions are produced by a large language model. The model is instructed to work only from the primary texts of each tradition, to cite passages by book, chapter and verse (surah and ayah, Ang, chapter number, or paragraph), to keep quotations short, and to note the translation family it is quoting. For Shinto, which has no single canonical scripture, it is instructed to say so and to point to the Kojiki, Nihon Shoki or norito where relevant — or to state that a theme is expressed through practice rather than text.
4. How citations are verified
On editorial pages, each citation was opened in the linked reference edition — Bible Gateway, Quran.com, Sefaria, SuttaCentral and Access to Insight, holy-bhagavad-gita.org, SikhiToTheMax, the Bahá'í Reference Library, the Chinese Text Project and Sacred Texts — and the wording was compared with the summary before publishing. If a passage could not be confirmed, it was not published. For live answers, the model is instructed never to invent, guess or approximate a reference; when it is unsure of an exact verse it must name the text without a verse number. Live citations are still machine-generated and should be opened and read before being relied upon. See Our Sources for the editions and Editorial Standards for who checks what, when pages were last reviewed, and how to report an error.
5. Neutrality safeguards
- Traditions are never ranked, scored or declared correct, and no answer recommends a religion.
- Every requested tradition receives its own section of comparable length; none is treated as the default against which others are measured.
- Descriptive language is used throughout (“the Qur'an describes”, “Theravada teachers hold”), not devotional or persuasive language.
- The same instructions apply to every tradition; there is no tradition-specific tuning that favours one over another.
6. How disagreements are handled
Disagreement between traditions is presented in the “Where they differ” section without adjudication. Disagreement within a tradition is named rather than smoothed over: when denominations, schools or lineages read a passage differently, the answer says so briefly. Where scholarship itself is uncertain — dating, authorship, or the meaning of a difficult text — the honest answer is that it is uncertain.
7. Limitations of AI
Live answers are generated by a language model and can be wrong: a passage may be summarized imprecisely, a nuance missed, a reference misnumbered, or a minority view presented as the majority one. The model cannot replace scripture, scholarship or a living community, and it does not know what is true. For that reason the Guide is designed as a well-organized reading list: every claim points to a text you can open, and the last step is always to read it.
The Scripture Guide uses artificial intelligence to help organize and explain religious texts and traditions. AI-generated responses may contain errors or omissions. When accuracy is important, consult the original scripture and authoritative sources.