AI & editorial policy

We use AI to help readers understand stories covered by many outlets at once. Everything AI writes or draws is labelled, linked to its sources and correctable. This page explains exactly what is automated and what isn't.

What is automated — and what is AI

• Not AI: collecting headlines from publishers' public feeds, sorting them into sections, placing them on the map, grouping reports about the same event, and balancing the front page. These are rule-based programs whose behaviour we can explain line by line.
• AI-written: Insights. When at least three outlets report the same event, a large language model writes a short synthesis: key points, a few paragraphs, key facts, and a note on how the outlets' accounts differ. Every Insight carries an AI Insight label and a notice at the top of the page.
• AI-drawn: illustrations. Insights may carry an illustration generated by an image model. Each is stamped “AI-generated illustration” on the image itself and in its description.

What the AI is given, and the rules it follows

The model receives only the headlines and short summaries that publishers distribute in their public feeds — never full articles taken from their sites — and it is instructed to:

• use only information in those reports, adding no outside facts, figures, names or quotes;
• cite the reports behind every point (the small numbers in the text link to them);
• write in its own words, never reuse a headline, and never quote directly;
• attribute claims to whoever made them, and say plainly where outlets disagree;
• stay neutral: no opinion, prediction or advice.

Automatic checks before anything is published

• Citations — any point without a valid source is removed.
• Copying — if any run of ten words matches a source, or the headline is close to a source's headline, the draft is rewritten.
• Figures — every number in the draft must appear in the source reports. A draft that still fails after one rewrite is held back for human review and not published.
• Spending and volume limits, so the system can't run away with itself.

Insights that pass these checks are published without a human reading them first. That is why we label them so clearly, link every point to the original reporting, and keep editors able to remove any Insight instantly. Insights are rewritten as a story develops; the time shown is when the current version was written.

Images

• We do not show publishers' photographs. Stories from other outlets appear with a generated tile showing the section, place and the outlet's icon.
• Our AI illustrations are deliberately symbolic and non-photographic. They never depict real, identifiable people, and are never presented as a record of an actual event.

Sources and balance

We carry outlets from every region, listed on the sources page, and the default front page caps how much any one country or outlet can take up. Headlines and summaries remain the property of their publishers, and every story links to the original.

Corrections

Email [contact@yourdomain] with the page link and the error. We will correct, regenerate or remove the Insight — usually within one working day — and where a correction changes the substance of an Insight we note it on the page. Publishers can ask us to change how their content is used, or to remove it, at the same address.

What we will not do with AI

• Publish AI content without a label.
• Invent quotes, or make realistic images of real people or events.
• Present an AI synthesis as original reporting — it is a summary of others' reporting, and says so.