This page explains exactly how stories, ratings and summaries on Kodawire are produced, including what is automated, what is reviewed by people, and where the method has limits.
Kodawire reads the public RSS feeds of roughly 370 English-language news publications in more than 70 countries. We store each article's headline, short description, publisher, link and publication time. We do not republish full articles: every story links to the original reporting, and headlines belong to their publishers.
Non-English articles are skipped, so coverage from publications that only publish in other languages is not represented.
Articles about the same event are grouped into a single story so you can compare outlets side by side instead of reading the same news ten times. Two articles are grouped only when their headlines share several distinctive words and at least one named person, place or organisation, and were published within about 48 hours of each other.
Generic words such as “election”, “court” or a shared year never count as a match on their own. When a pair is borderline, an AI model is asked whether the two headlines describe the same event; if it is not confident, the articles stay separate.
Grouping is automatic and can still be wrong. If you see an article that does not belong in a story, tell us at hello@kodawire.com and we will fix it.
Every publication is placed in one of three buckets: Left, Center or Right. This rating describes the outlet's usual editorial lean, not the truth of any article.
A story's coverage split (for example “20% Left · 70% Center · 10% Right”) is simply the share of grouped articles that came from outlets in each bucket. It is not a measure of the article text itself.
Each outlet carries a factuality rating of High, Mixed or Low. A story's credibility score (0–100) is the average over the outlets covering it, where High counts as 89, Mixed as 78 and Low as 54, kept within 55–95. The score is only shown when at least two sources cover the story.
The score describes the track record of the outlets involved. It is not a fact-check of the specific story, and a high score does not mean every claim in the coverage is accurate.
Kodawire uses large language models from third-party providers for:
AI text is generated only from the articles linked on the story. Summaries that do not match the story's headline are hidden automatically. AI summaries are not individually reviewed by a person.
Editor's notes.Some stories carry a signed Editor's note with added context. An AI model may draft the note from the linked articles; a named editor then checks it against the sources, rewrites what is wrong and approves it. Notes produced this way say “Drafted with AI assistance; reviewed and approved by” the editor. Notes written entirely by an editor show only their name.
A story is offered to search engines only when it is a genuine comparison: at least three sources from at least two of the Left / Center / Right buckets, a summary that matches the headline, and not a transfer rumour, celebrity gossip, analyst price-target note, live blog or listing. Other stories stay available to readers but are marked noindex.
Every story is re-checked on a schedule, including older ones. Articles that turn out to describe a different event are removed from the story and regrouped, source counts and the Left / Center / Right split are recalculated from the current outlet ratings, and any summary that no longer matches the headline is removed and regenerated. Stories left with no articles are archived.
Trending ranks stories in the selected window by reader activity on Kodawire: page views ×1, story card downloads ×3, comments ×4 and shares ×5.
If a rating, grouping or summary is wrong, email hello@kodawire.com with the story link. Ratings of reviewed outlets are updated when we change them; this page is updated whenever the method changes.
