Skip to content
AIRanking

Trust

How the ranking works

How AIRanking scores tools: upvotes, Bayesian editor ratings, recency, and editorial weight. Seed metrics are estimates, not paid placements.

Every score is computed at build time and baked into HTML. Visiting a page never queries the database. The formula is identical in TypeScript and in the Supabase ranking function.

score = 100 × (0.45 × ln(1+upvotes)/ln(1+max_upvotes) + 0.25 × Bayesian rating + 0.15 × recency + 0.15 × editorial_weight/10)

  • Upvote counts and star ratings shown on tool pages are editorially seeded estimates for the directory launch — they are not live user vote totals or verified review aggregates. After the upvote Worker ships, community votes can update totals on a rebuild cycle.
  • Editor ratings use a Bayesian prior (weight 20, mean 3.8) so a handful of marks cannot dominate; the displayed rating is still an editorial seed, not a crowdsourced score.
  • Recency decays with an 18-month half-life from last verified / launch date.
  • Editorial weight (0–10) captures product quality the numbers miss, and is disclosed here rather than hidden.

Affiliate URLs, when set, use rel="sponsored noopener". Inclusion is not for sale; Fast-track listing is a future workflow, not a ranking boost.