How we use AI
What the AI does
Groups the news
AI sorts reports from many outlets into topics, so one story is not counted ten times.
Pulls out the facts
It extracts facts, numbers and attributed statements from published reports, each tied to the report it came from.
Drafts the brief
It writes the brief in plain language from those collected facts only. It is not allowed to add facts of its own.
Automatic checks
Before anything is published, a brief is rejected when:
- a number in the text does not match a number in the sources or the data;
- a sentence has no source, or a timeline date is not supported;
- sentences copy the source reports instead of summarising them;
- the wording takes a side or judges people.
When an editor steps in
Every brief waits in a review queue, where an editor reviews, corrects, publishes or rejects it. When we switch it on, an AI editor may also publish briefs that passed every check, including sensitive ones. A brief that fails a check always waits for a person. The line under each headline tells you which happened.
Records, year-on-year changes and rankings are calculated by our software from official data, not written by the AI.
The AI models we use are Google Gemini, with Groq-hosted open models as a backup.
Read the latest briefs