How Bramble works

News about the same event is scattered across outlets that frame it very differently. Bramble puts one event's coverage side by side, writes a neutral account that every sentence of can be checked, and shows how outlets on the left, in the centre and on the right each tell it.

1. Reading

Every few minutes we read the public news feeds of the outlets listed on our sources page, and fetch each article where the outlet's robots.txt allows it. Article text is kept privately for analysis; we only ever show short quotations, always credited and linked.

2. Understanding

An AI newsroom editor (Anthropic's Claude) files each article into a story, one real-world event, and records who and what it is about: people, organisations, parties, places, laws. These form a knowledge graph that connects stories over time, and records the factual claims each outlet makes and whether it reports, supports or disputes them.

3. Writing the neutral account

Once at least 2 outlets have covered a story, Claude reads coverage from across the spectrum and writes a neutral account. Every sentence must cite the articles it rests on with exact quotations, and our software checks that each quotation really appears in the cited article before anything is published; if a quote can't be found, the account is rejected and rewritten.

4. Comparing the coverage

Alongside the neutral account we show what every side reports, where accounts differ, how each side frames the story, what it emphasises and leaves out, the charged language it uses (quoted exactly), and how each person or organisation involved presents their own position. The count of left, centre and right outlets comes from the outlets' bias labels, not from the AI.

Limits

AI can misread and misfile. Editors review the work, and you can always check a sentence against its sources, which is the point of citing them. Bias labels are generalisations about outlets, and a neutral account is a best effort, not the last word.