How it works

There is no editorial curation here and no personalised engagement ranking. Everything below is deterministic: the same inputs always produce the same feed.

1. Clustering headlines into stories

Different outlets describe the same event in different words and different languages, so there is no shared identifier in the raw data. We derive one from the headline: strip accents and stopwords, resolve known entities to a canonical form (so "Estados Unidos", "United States" and "Vereinigte Staaten" collapse together), sort what's left, and use that as the story's key.

Sorting matters — it makes the key order-independent, so two headlines that use the same significant words in a different arrangement land in the same cluster. A second pass then merges near-duplicate clusters, and any cluster an editor merges or splits by hand is locked against further automatic changes.

2. Scoring importance

Each cluster is scored as the product of five signals rather than a sum, so a story has to do well on several at once — being loud on one alone doesn't carry it:

• Position decay — where the story sat on each outlet's front page. Front pages are already editorially ordered, so this reads each paper's own judgement instead of collapsing to whatever we ingested most recently.
• Corroboration — how many distinct outlets, across how many countries, independently ran it. This is the dominant term.
• Authority — the publishing outlets' quality tier, from global wire services and papers of record down to local outlets.
• Velocity — how fast coverage is accelerating, so a story breaking now outranks one that has been steady for two days.
• Category weight — a light correction so a heavy sports day doesn't crowd out everything else.

3. Outlet tiers

Every publisher we've seen sits in the outlet registry at one of four tiers, and a newly discovered outlet starts at tier 3 until someone forms a view on it:

• Tier 1 — elite global wire services and mastheads with genuinely global reach. Deliberately narrow.
• Tier 2 — major national papers of record, public broadcasters and national wires: high quality or high reach within their own country or region.
• Tier 3 — the default, and where every newly auto-discovered outlet starts.
• Tier 4 — local and small-market outlets.

4. Blindspots

Two different questions share the name. Worldwide, a blindspot is a story big in one country's press that has no international pickup at all. Scoped to a country, it's the inverse: stories that several other countries' press are running and that this country's press has not touched.

Both fall out of the same coverage data — for any story we know which countries have run it and which countries' press is currently active but silent. Silence is only meaningful against a recent window, so a country that has published nothing at all lately isn't counted as staying quiet on anything.

Known limits

Worth stating plainly. Section classification is a fixed multilingual keyword table, not a per-article model, so it misfiles edge cases. The city gazetteer covers major cities, so a headline naming a smaller town goes undetected. Clustering keys on headlines alone, so two genuinely different stories that share most of their significant words can merge. And an outlet that syndicates a wire story verbatim counts as corroboration even though it did no independent reporting — the single biggest known weakness in the ranking.

Frequent questions

Does NewsOverlap write any of its own news?

No. It indexes headlines and links out to the publishers who wrote them. The original work here is the analysis layer — the clustering, the corroboration ranking, and the coverage-divergence measurement — not the reporting.

Does NewsOverlap rate outlets for political bias?

No. Outlets carry an optional stance label where one is well established, but ranking uses only the quality tier and never the stance. Nothing is boosted or suppressed for its politics.

Why does a story show as uncovered in a country that did cover it?

Coverage is measured over the outlets and feeds we actually index, in a recent time window. If a country's coverage came from an outlet not in the index, or fell outside the window, it reads as silence. It is a measure of observed coverage, not proof of absence.

Is my reading history tracked?

There is no account requirement and no reading history. Choosing countries stores a preference against an anonymous id in a cookie; it is not tied to an identity and there is no per-article tracking.