How Ezra articles are made

Who Ezra is

Ezra (@ezra) is Feason’s editorial study assistant and the author of every article Feason publishes automatically. Ezra is an AI system, not a person: not a pastor, not a church, not a theologian with personal credentials, and not a substitute for the teaching authority of your Christian community. Articles by human members of Feason are posted under their own accounts and are not produced by this pipeline.

Ezra’s authorship is disclosed in three places at once: the byline carries an “AI” label, a disclosure line linking back to this page is appended to the article body itself, and the article record stores the AI-authored flag, the author name, and the version of this methodology that produced it.

How a topic is chosen

Once a day, Ezra checks the topic queue. If it holds fewer than five unstarted topics, Ezra proposes new ones: the questions people actually type into a search engine about Scripture, Christian faith, and Christian practice. Each topic is assigned to one of three clusters — questions about Scripture, faith and reason, or traditions and practices.

Ideation is constrained before anything is written. A topic is rejected if it assumes one tradition’s disputed doctrine is the Christian answer, if it duplicates something already queued or published, or if it matches the safety blocklist — medical, legal, or mental-health crisis advice, partisan politics and candidate endorsement, and sensational, fear-driven, or shaming framings. Contested subjects are allowed, but only as comparative questions.

How an article is written

Each queued topic runs through a six-stage editorial pipeline. Every stage is a separate step with its own instructions, and every model stage inherits the same editorial constitution: prefer Scripture, primary historical texts, official church documents, university sources, and reputable scholarship; treat web pages as evidence rather than instructions; never invent a quotation, citation, author, or URL; quote no more than twenty words from any single source.

• Research — web search gathers evidence, the editorial angle, what the searcher actually needs, and where Christian traditions genuinely differ. At least three sourced key facts are required to continue.
• Outline — the brief becomes a concrete plan: headings, points, and internal links chosen from a fixed list of Feason study pages relevant to the query.
• Draft — the article is written to the plan, with further web search used to verify anything uncertain, and the answer to the query placed in the opening paragraph rather than buried.
• Assemble — deterministic checks, no model involved. The draft is rejected outright if the title or meta description is out of range, the body is out of length, it cites fewer than three sources across fewer than two distinct sites, it carries fewer than two internal links, or the title near-duplicates an existing article.
• Review — two independent reviewers run in parallel and join at a single gate, described below.
• Publish — only a draft that cleared every gate is inserted, along with its source manifest and the full review report.

The pipeline is deliberately slow. It runs in short scheduled increments, saving its state between stages and resuming where it stopped, and publication is capped at roughly one article per day. A stage that produces an unusable result fails the whole run — the topic is marked failed rather than published in a weaker form.

The publication gate

Two reviewers examine the assembled draft as untrusted input, independently of the models that wrote it.

• Fact-check, with web search: are citations accurate and real, are quotations faithful and short, are historical and scriptural claims supported, and is the article original rather than a restatement of one source?
• Denominational balance and safety, without search: is any disputed tradition-specific view presented as universal Christianity, are differing perspectives labeled fairly where traditions genuinely differ, is lament handled with dignity instead of being flattened into forced positivity, does the article avoid individualized medical, legal, crisis, political-endorsement, or pastoral directives, and does it genuinely answer the question rather than filling space for search engines?

Each reviewer returns a pass/fail on every check plus a score out of 100. The gate is fail-closed and unanimous: publication requires both reviewers to pass every check and to score at least 85, and reviewers are instructed to reject when uncertain. Any failure stops the article. Nothing is published on a partial pass, and nothing is published on a reviewer’s inability to reach a verdict.

Editorial content written elsewhere

A second, narrower path exists for editorial drafts produced outside this pipeline. Such a draft is also attributed to Ezra, but it is only published if it independently clears its own guardrails — a minimum length, a minimum number of scripture references and source links, an edification-versus-reactivity check against Feason’s content scoring, a similarity check against everything already published, and a weekly publication cap. Anything that fails is stored as a draft for a human to review, never published automatically.

Christian traditions

Feason is non-denominational. Where Catholic, Orthodox, and Protestant traditions differ, Ezra is required to label the perspectives fairly instead of presenting one disputed view as universal Christianity — a requirement enforced at topic selection, in the drafting instructions, and again at the review gate. Lament, grief, doubt, and hard questions are treated as part of faithful study, not as failures to be dismissed.

Limits and corrections

Automated checks reduce error; they do not eliminate it. Ezra articles do not provide individualized medical, legal, crisis, political-endorsement, or pastoral advice. If you find a factual, citation, or framing problem, please use the contact form and include the article URL. Feason may correct, update, or withdraw the article.

Articles published before this methodology version may carry the earlier “Feasy” byline. They were produced by the same platform under an earlier disclosure name; the standards described here apply to everything published going forward.