How we use AI

We use software, including AI language models, to help a small newsroom follow a very large amount of news. Here's exactly how.

Grouping and sorting

Software groups reports of the same story from different outlets and sorts stories into sections and topics.

Our articles

Our stories are written from the full reports that other outlets published, which we credit and link on every story. An AI model that runs on our own servers drafts the story in our words, under strict rules: only facts that are in those reports, names and figures exactly as reported, every claim attributed, no quotes that the reports don't contain. Software then checks every draft against the reports and throws it out if it finds a number, name or quote that isn't in them, or a sentence copied from them. A second pass, also by AI, checks the draft against the sources, and a story that still fails is held back. The short summaries on our home page and the Hindi summaries are written from the checked article. Every article says it was written with AI assistance. Where we can't read a full report, we don't write the story: we show the headline and link to the outlets.

What AI doesn't do

AI doesn't decide what's sensitive: our editorial rules and editors do, and we don't write AI articles on sensitive stories. AI doesn't make images of real people or events for us. Our live anchor's voice is generated with the presenter's consent.

Mistakes

AI can get things wrong. If you see an error, please tell us through our corrections page and we'll fix it quickly.