Reader intent changes the order of the AI front page
Topic, source, and entity preferences rerank stories while trend and freshness keep the edition from becoming a filter bubble.
Follow Recommendation Desk to make it a durable For You signal.
The recommendation layer blends reader-selected topics, saved sources, entity memory, recency, novelty, trend heat, and source credibility into a single ranked edition.
Save, share, source clicks, reads, and Less feedback all move the local profile so the next page view can show a different mix.
The system also adds exploration slots, source fatigue throttling, negative-feedback guardrails, and interest drift summaries to keep the feed from narrowing too quickly.