Two scores are stored for every listing. Payment never changes either of them.
The default table sorts by the overall score. AI Flood reviews are a recency-weighted average: a recent review counts more than an old one. When a public web score exists, it is blended in and never falls below 30% of that blend, so our reviews are capped at 70%. With no web score, a thin sample is still pulled toward the mean for its kind (AI tools are not mixed with models). The table shows the blend, plus "AI Flood users" and "Across the web" when each exists. Comments are not part of this score.
Heat is recent engagement: ratings, views and verified change events, each with exponential time decay. It is recomputed every 30 minutes, blended with the previous value so a single spike cannot fling a listing up the board, and a listing can move only a limited number of places on each update. Heat orders the Hot strip and the Trending tab, and it is the 7-day arrow. Until there is any engagement, the strip shows what was published in the last seven days.
AI tools, models, services and courses share one listing table and one attribute store. Each kind has its own ranking. A new category is a row in the attribute definitions, not a schema change. Geography is either worldwide or a list of countries. For and Use tags describe who an AI tool is for and what it does.
A sponsored placement is pinned above a ranking and labelled. The ranking code does not read the sponsorship tables. A sponsored product still keeps whatever organic rank its scores earn.
