How we use AI
We believe readers are owed a plain account of how their news is made. Here is ours.
What the AI does
Zepy is produced by an AI-assisted newsroom, built on large language models from Anthropic (Claude) and Google (Gemini). Separate agents act as a researcher, a reporter, a standards editor, an SEO editor and a photo editor:
- The researcher searches the web for reliable reporting on a topic and collects sources.
- The reporter writes a draft that attributes each claim to those sources.
- The standards editor checks the draft: every quotation is matched word for word against the source text, claims are compared with the sources, and the piece is screened against our editorial scope.
- The photo editor looks for an openly licensed image and records its credit.
Our short briefings work differently. They summarise the headlines and summaries that outlets publish in their own public feeds, entirely in our own words. Every item names the outlet and, where the feed gives one, the reporter, and links to the original story. Briefings never reproduce an outlet’s text or photographs.
What the AI does not do
- It does not invent quotations, sources, people or statistics.
- It does not write about subjects outside our editorial scope; those are held for human review.
- It does not publish under a false claim of being a human journalist. Bylines are editorial identities of the newsroom, and each article is labelled AI-assisted.
Human oversight
A person owns the newsroom and sets its rules. Stories flagged as sensitive and any story that fails an automated check are held for a person to review before they can be published. Other stories can be published automatically once they pass the checks, and they remain labelled as AI-assisted.
Mistakes
AI can be wrong. If you find an error, tell us and we will correct it publicly.