Trust & methodology
Editorial Methodology — How Content Gets Published Here
How BestAIAgent.in actually decides what to publish: the real two-tier verification model, what editorial pages are held to, and what gets rejected.
This page documents the actual publication process on this site, matching CONTRIBUTING.md and the real code in lib/catalog.ts rather than an aspirational editorial-board process that was never built.
Two tiers, not a review board
Every entity starts in a legacy/imported pool with just a name and a source link — reachable, but noindex. It's promoted to a verified, indexable entity only by adding a real Entity record plus a matching EvidenceRecord built from an actual API fetch (see /trust/evidence-methodology). There's no separate named-reviewer sign-off step distinct from that promotion — the evidence record itself is the review artifact, and anyone can check it by re-fetching the cited source.
What editorial (non-entity) content is held to
Authority pages, glossary entries, and guides don't go through the entity evidence gate — they're editorial by default. What is checked:
- No fabricated statistics, ratings, or "trusted by" claims. If a number isn't sourced, the plain-language version is used instead of inventing one.
- No compliance or certification claims about a named company without a real source (see /trust/privacy-dpdp-editorial-policy for the sharpest version of this rule).
- Internal links actually resolve — checked with a real grep-based script before merging a batch of new pages, documented in
CONTRIBUTING.md. - No mass-generated, name-swapped pages published purely to rank for a keyword variant.
What we refuse to publish
Entities without a real, checkable evidence receipt stay in the noindex tier — no exceptions for coverage or completeness. Comparisons whose title doesn't match what their a/b fields actually point to don't ship (see /trust/comparison-methodology for the real incident that taught us to check this). Schema.org blocks asserting something the page doesn't actually show — fabricated AggregateRating, invented offers.price — are rejected in review.
Related methodology pages
- Evidence methodology
- Rating methodology
- Comparison methodology
- Source classification
- Corrections
- Affiliate disclosure
- Author attribution
Source receipts
- Internal —
CONTRIBUTING.md. Documents the real two-tier model, the promotion workflow, and the specific mistakes (hand-typed hashes, mismatched comparison titles) this policy is designed to prevent. - Internal —
lib/catalog.ts(isEntityIndexable,isPublicIndexableComparison). The real gate functions.
Last reviewed
- Page last reviewed: 2026-09-02