Methodology

Primary source: the GDELT Project

Geotone's core signals derive from the GDELT Project, which monitors news media worldwide in 65+ languages and publishes machine-readable data files every 15 minutes. We download those files — the Global Knowledge Graph article stream (English and translated editions) plus the event and mention tables — on a roughly two-hour schedule, process each 15-minute window exactly once, and store only daily aggregates. Geotone never stores or republishes article text; article listings on topic pages are headline, source domain, and link only, pointing to the original publisher.

Coverage volume

For each topic we hold a fixed, published query (shown on request) and match it against every article title in GDELT's 15-minute files, using the same grammar GDELT uses (AND/OR grouping, exact-phrase quotes, whole-word matching). The same files give the total number of articles GDELT monitored in each window. We sum both across each UTC day:

coverage share (%) = 100 × matched articles ÷ all monitored articles

For countries, volume counts articles whose extracted locations include that country (GDELT's location country codes), measuring coverage about the country as a share of all monitored coverage — how much the world's press is talking about a place, not how much its own press publishes. (Before 2026-07-03 this series measured press output from each country; the two scales are not directly comparable.)

Tone

Tone is GDELT's standard document-tone measure: positive words minus negative words as a share of all words, roughly bounded -100 to +100 and typically between -10 and +10. For each topic or country we average the tone of its matching articles across each UTC day. Values below about -2.5 read as negative coverage; below -5, strongly negative.

Week-over-week change

WoW (%) = 100 × (mean of latest 7 days − mean of prior 7 days) ÷ mean of prior 7 days

Shown only once a full fortnight of data exists. Changes within ±2% are described as "steady" in page summaries.

Surge detection (z-score)

z = (latest daily volume − mean of trailing 28 days) ÷ sample std-dev of trailing 28 days

The latest day is excluded from its own baseline. A topic or country is marked surging when z ≥ 2 — meaning today's coverage volume is at least two standard deviations above its own recent norm. Z-scores require at least 8 days of history and are suppressed when variance is zero.

Update cadence

• Topic and country volume and tone: recomputed roughly every 2 hours from GDELT's 15-minute files.
• The wire: reclustered on the same 2-hour cycle; story timelines extend as chains grow.
• Article samples: refreshed daily per topic, capped at 40 links per page.
• Pages re-render as new data lands; the "Data updated" line shows the newest datapoint's date.

Event indexes & directed relations

Country pages carry daily counts of CAMEO-coded events geolocated to the country, from GDELT's event stream: protests (CAMEO root 14), diplomacy and cooperation (04–05), and coercion through mass violence (17–20). Relation pages additionally split events by direction — who initiated actions toward whom — with the mean Goldstein score (−10 most conflictual to +10 most cooperative) of those events. Event coding is GDELT's, automated and imperfect; counts indicate activity levels, not verified incident tallies.

Emotional measures

Topic pages show emotion densities from GDELT's Global Content Analysis Measures (GCAM): occurrences of anxiety-, anger-, and positive-emotion-related words per 1,000 words of matching coverage, using the LIWC dictionaries (Pennebaker et al.). English-stream articles only. Where tone says whether coverage reads negative, these say what kind of negative. Like tone, they are lexical measures — trends matter more than levels.

Reader interest

Entity pages and mapped topic pages show daily human pageviews of the corresponding English Wikipedia article (Wikimedia Analytics, bot traffic excluded). Coverage volume measures what the press supplies; pageviews measure what readers seek out — divergence between the two is its own signal. Topic-to-article mappings are curated and disclosed; topics without a single defensible article carry no reader-interest module.

Attention gaps (underreported)

The underreported page lists significant disasters (GDACS Orange/Red alerts; USGS M5.5+ earthquakes) from the last 14 days beside the affected country's measured coverage response: mean coverage share from the event day onward versus the country's own trailing 14-day baseline, as a percent change. “Muted” flags a severe event whose response stayed under +25%. Severity scores come from the source agencies; the gap is a measurement of attention, not a judgment of newsworthiness.

The wire & story timelines

The wire clusters the last 48 hours of headlines into stories (grouping syndicated copies of the same report), ranks each story by a significance score — how many independent, authority-weighted outlets carry it, boosted when it matches a tracked topic, carries consequence themes, or links to coded conflict events — and threads related stories into timelines. The scoring recipe and the source authority list are published on the wire page itself.

Known limitations

• GDELT measures media attention, not ground truth. A surge means the press is writing more, which correlates with — but is not identical to — events on the ground.
• Country volume reflects global media attention, not importance: places with a large English-language wire presence are over-represented, and coverage of small countries can be dominated by a single event.
• Tone is a lexical measure. Sarcasm, translation artifacts, and topic-inherent negativity (war coverage is always negative) mean trends matter more than levels.
• GDELT's 15-minute files occasionally arrive late; missed windows are caught up on the next cycle, so the newest datapoint can trail real time by a few hours.

Licensing & reuse

Geotone's derived aggregates (daily volume, tone, z-scores) are licensed CC BY 4.0— cite "Geotone / GDELT". Headlines and article links remain the property of their publishers. Enrichment data comes from Wikidata (CC0), Wikipedia (CC BY-SA, always attributed inline), the World Bank, UN OCHA ReliefWeb, and UCDP. Questions? See about & contact.