When a new article arrives, AI reads its headline and summary to assign one or more topic labels (like “World” or “Technology”) and identify which countries the article is about. This is a categorisation task — the AI does not rewrite or summarise the article.

Within each topic, AI groups articles that cover the same underlying event. It also generates a short, neutral title and summary for the cluster. When new articles arrive about an existing story, they are merged into the existing cluster rather than creating a duplicate.

AI scans recent story headlines to find specific events or entities that are being widely covered (e.g. “UK Budget” rather than a generic “Politics”). This powers the trending bar in the site header.

For multi-source stories, AI compares the headlines and summaries from different outlets to produce a structured analysis: what the sources agree on, where they diverge, and what perspectives may be absent. For political topics, it also maps outlets on a left-right spectrum using our database of known outlet leanings (which are editorially assigned, not AI-generated).

The analysis is cached — it is only regenerated when new sources are added to a story, not on every page view.