News Intelligence API & Scraper
The News Intelligence API returns news search and analysis as clean JSON.
🤖 Using an AI assistant? Copy this link into ChatGPT / Claude / Cursor — it reads every endpoint and parameter instantly and tells you if this API fits your use case.
The primary search endpoint returns articles with title, URL, source domain and country, published date, language, snippet and image, and you can pull a timeline, sentiment, entities, trending, top_headlines and by_source. It is built for media monitoring, market intelligence and research that need structured news data with analysis layers. One ReefAPI key, one shared credit pool, the standard envelope.
Which upstream answers which action, and what a busy window costs you
news-intel merges three very different sources into one article shape, and which one answered decides which fields are populated. GDELT allows roughly one request every five seconds and that window is shared, so it is routinely unavailable. Actions that can fall back to Google News still return ok true with fewer fields, while actions that cannot fall back return an error instead of a short answer. meta.sources_used and meta.source_errors tell you which case you got.
| Action | Backed by | What you get, and what you lose |
|---|---|---|
| search, coverage, by_source, batch | Google News RSS plus GDELT, merged (merge_google_news defaults true) | on a busy GDELT window the call still succeeds on Google News alone, and language, snippet, image, source_country and tone all come back null |
| timeline, sentiment, entities | GDELT only | no fallback exists, so a busy window returns UPSTREAM_HTTP with a retry-in-a-few-seconds message rather than an empty series |
| top_headlines | Google News section feed | topic is a hard enum of WORLD, NATION, BUSINESS, TECHNOLOGY, ENTERTAINMENT, SPORTS, SCIENCE and HEALTH, and an unlisted value is rejected |
| tech | Hacker News top, new or best, or Algolia full-text search | the only action that returns the publisher's own URL, plus points, author, num_comments and hn_url |
| sources | a static directory, no upstream call | 19 US outlets, editions DE ES FR GB HK IN QA US, and 12 language codes; by_source itself accepts any publisher domain |
| trending | GDELT volume buckets | six default buckets: breaking news, world, tech, business, politics and science |
Google-News-backed articles carry a news.google.com/rss/articles/ redirect URL rather than the publisher's link, so source_domain and publisher are the reliable identity fields, and only tech returns direct publisher URLs. trending is GDELT-only but does not error when GDELT is busy: it returns ok true with volume 0 and an empty sample_headlines for every bucket, so check volume before you use it.
Real request and response JSON
Captured from the indexed primary action, search, on .
{
"method": "POST",
"url": "https://api.reefapi.com/news-intel/v1/search",
"headers": {
"x-api-key": "$REEF_KEY",
"content-type": "application/json"
},
"body": {
"query": "OpenAI",
"timespan": "2d",
"max": 40
}
}{
"ok": true,
"meta": {
"api": "news-intel",
"endpoint": "search",
"mode": "live",
"latency_ms": 3220.1,
"record_count": 97,
"bytes": 129115,
"cache_hit": false,
"method": "public_news_api",
"freshness": {
"google_news": "near-live (minutes)"
},
"total_before_dedup": 104,
"removed_duplicates": 7,
"sources_used": [
"google_news"
],
"query": "OpenAI",
"timespan": "2d",
"source_errors": [
"gdelt: throttled (5s window)"
]
},
"data": {
"articles": [
{
"title": "Apple Sues OpenAI, Alleging It Stole Trade Secrets",
"url": "https://news.google.com/rss/articles/CBMiZEFVX3lxTE5XaDJLOUlqNlc1NW9HckNySzZyWFQzWWF3ZDNXXzJMdE95ck12cEYyck91SGVVeVRYWWVTWUdhTlNuTXFBM1lsX2J1TS1LcGN5MzNiMkVGN0c0cUxhUXBOVktGU1I?oc=5",
"source_domain": "wsj.com",
"source_country": null,
"published_at": "[redacted-phone]T07:00:00Z",
"language": null,
"snippet": null,
"image": null,
"source": "google_news",
"tone": null,
"publisher": "WSJ",
"duplicate_count": 0,
"also_reported_by": []
},
{
"title": "China's Moonshot AI claims Kimi K3 can rival OpenAI and Anthropic",
"url": "https://news.google.com/rss/articles/CBMiWkFVX3lxTFBRRWRCTzZXLWRlVV9jYTRzN3ktc1lFLWExbVlPa09HSVEwR0VyVWhxYkcxNG5mRUVHcDdGRVFaQ0hoOU95OHQ5ZmtRWkYxcnh6Z2dkYUowZ0FmQQ?oc=5",
"source_domain": "bbc.com",
"source_country": null,
"published_at": "[redacted-phone]T16:53:04Z",
"language": null,
"snippet": null,
"image": null,
"source": "google_news",
"tone": null,
"publisher": "BBC",
"duplicate_count": 0,
"also_reported_by": []
},
{
"title": "AI’s Wider Availability Is Good for China, Not Great for OpenAI and Anthropic",
"url": "https://news.google.com/rss/articles/CBMif0FVX3lxTFBSeWJBNG5TTE9zVV9URzRmX1huMWJkSW11Q2J4azZ0WXpVczR6Z19yMkFjNU1ZNXgxWjNfYkFETk1QZUR0LVpLYkhhRDFvV3FPWFlnT1pvTmdKQVJBOEU1UzFHLWViQ29zRlM3ckVqMVRDRnp3dlhJbW1uc21wLUU?oc=5",
"source_domain": "wsj.com",
"source_country": null,
"published_at": "[redacted-phone]T13:38:00Z",
"language": null,
"snippet": null,
"image": null,
"source": "google_news",
"tone": null,
"publisher": "WSJ",
"duplicate_count": 0,
"also_reported_by": []
}
],
"query": "OpenAI",
"timespan": "2d",
"total_before_dedup": 104,
"removed_duplicates": 7,
"sources_used": [
"google_news"
],
"freshness": {
"google_news": "near-live (minutes)"
}
}
}What the News Intelligence API does
| Action | Description | Concrete use case | Key params |
|---|---|---|---|
| search | Global news search across GDELT (+ optionally merged Google-News-RSS), de-duplicated into one unified article shape. The core 'what's the news on X' call. Filter by timespan/language/country/sort; each article carries title, url, source_domain, published_at, language, snippet, image, duplicate_count, also_reported_by[]. | Content platforms call search to get global news search across GDELT (+ optionally merged Google-News-RSS), de-duplicated into one…. | query, timespan, start, end, language, ... |
| timeline | How much is X in the news over time — GDELT volume-over-time (timelinevol) for a query as [{date, value}], with peak detection (max point + average + latest). SLOW BY NATURE: the analytics index behind this endpoint currently answers in about 20-25 seconds and refuses a large share of connections, so expect a ~20-25s call and an occasional honest 'source did not answer' rather than a sub-second reply. If the source misses, a reading up to 30 minutes old may be returned instead, always labelled with stale_age_seconds. For fast news use search / top_headlines / by_source (~1s). | Research tools call timeline to get how much is X in the news over time. | query, timespan, start, end, language, ... |
| sentiment | Sentiment / tone analysis for a query — GDELT tonechart gives the distribution of article tone (how positive/negative coverage is) as a histogram, a count-weighted average tone, a positive/neutral/negative breakdown, and the single most-positive and most-negative representative article. Answers 'how is X being covered — favorably or critically?'. (alias: tone) SLOW BY NATURE: the analytics index behind this endpoint currently answers in about 20-25 seconds and refuses a large share of connections, so expect a ~20-25s call and an occasional honest 'source did not answer' rather than a sub-second reply. If the source misses, a reading up to 30 minutes old may be returned instead, always labelled with stale_age_seconds. For fast news use search / top_headlines / by_source (~1s). | Community analysts call sentiment to get sentiment / tone analysis for a query. | query, timespan, start, end, language, ... |
| entities | Top entities for a query — the people/organizations and locations most mentioned in coverage of a topic, ranked by mention count. Organizations/people are mined from article titles; locations come from GDELT's source-country distribution. Metadata-only. Useful for 'who and where is this story about'. SLOW BY NATURE: the analytics index behind this endpoint currently answers in about 20-25 seconds and refuses a large share of connections, so expect a ~20-25s call and an occasional honest 'source did not answer' rather than a sub-second reply. If the source misses, a reading up to 30 minutes old may be returned instead, always labelled with stale_age_seconds. For fast news use search / top_headlines / by_source (~1s). | Media monitors call entities to get top entities for a query. | query, timespan, start, end, language, ... |
| trending | What's big in the news right now — a curated multi-bucket roll-up (breaking/world/tech/business/politics/science by default, or your own queries) ranked by recent volume, each with a few representative headlines. SLOW BY NATURE: the analytics index behind this endpoint currently answers in about 20-25 seconds and refuses a large share of connections, so expect a ~20-25s call and an occasional honest 'source did not answer' rather than a sub-second reply. If the source misses, a reading up to 30 minutes old may be returned instead, always labelled with stale_age_seconds. For fast news use search / top_headlines / by_source (~1s). PARTIAL BY DESIGN: each bucket is fetched independently, so a call can return fewer topics than requested; meta.bucket_errors names the ones that missed, and a repeat call fills them in (buckets already fetched are reused). | Content platforms call trending to get what's big in the news right now. | queries, timespan, language, country, max |
| top_headlines | Editor-curated top headlines by Google-News topic section (WORLD/BUSINESS/TECHNOLOGY/…) and/or country edition. De-duplicated. The 'front page' call. | Research tools call top_headlines to get editor-curated top headlines by Google-News topic section (WORLD/BUSINESS/TECHNOLOGY/…) and/o…. | topic, country, max, dedup |
| by_source | Recent articles from a specific publisher domain (reuters.com, bbc.com, …). Google-News site:<domain> is the reliable primary (near-live, ~100% on-source); GDELT domain:<domain> enriches with direct publisher URLs + tone when its 5s window is free (skipped silently otherwise — never blocks). Source-exact. merge_google_news=false → GDELT-only exact. | Community analysts call by_source to get recent articles from a specific publisher domain (reuters.com, bbc.com, …). | source, query, timespan, language, sort, ... |
| coverage | Full-story coverage — for a story or topic, discover which news outlets are covering it. Returns the most-covered stories first, each expanded to every outlet that carried it (outlet_count + outlets[]). Ideal for media monitoring and PR: see how widely a story was reported and by whom. | Media monitors call coverage to get full-story coverage. | query, timespan, start, end, language, ... |
| sources | Reference directory of supported news outlets, filterable by category (e.g. TECHNOLOGY, SPORTS) and country. Also returns the topic, language and country reference lists accepted by other actions. Note: by_source accepts any publisher domain — this directory is a curated starting set, not a limit. | Content platforms call sources to get reference directory of supported news outlets, filterable by category (e.g. | category, country |
| tech | Hacker News tech vertical — top/new/best stories, or a full-text Algolia search. Each story: title, url, points, author, num_comments, created_at, hn_url. | Research tools call tech to get hacker News tech vertical. | query, list, max |
| batch | Run up to 10 search queries in one call (shared concurrency, GDELT 5s-window paced). results[] preserves input order; each is a de-duped search result. | Community analysts call batch to get run up to 10 search queries in one call (shared concurrency, GDELT 5s-window paced). | queries, timespan, language, country, max, ... |
Call search from your stack
curl -X POST https://api.reefapi.com/news-intel/v1/search \
-H "x-api-key: $REEF_KEY" \
-H "content-type: application/json" \
-d '{"query":"OpenAI","timespan":"2d","max":40}'import requests
r = requests.post(
"https://api.reefapi.com/news-intel/v1/search",
headers={"x-api-key": REEF_KEY},
json={
"query": "OpenAI",
"timespan": "2d",
"max": 40
},
)
print(r.json()["data"])const res = await fetch("https://api.reefapi.com/news-intel/v1/search", {
method: "POST",
headers: {
"x-api-key": process.env.REEF_KEY,
"content-type": "application/json",
},
body: JSON.stringify({
"query": "OpenAI",
"timespan": "2d",
"max": 40
}),
});
const { ok, data, meta, error } = await res.json();Ask your MCP-connected assistant: call reefapi.news-intel.search with {"query":"OpenAI","timespan":"2d","max":40}.Who uses this API and why
- Brand-monitoring tools call search and sentiment to track coverage volume and tone.
- Analysts use timeline and entities to map how a story developed and who was involved.
- Dashboards use trending and top_headlines to surface what is breaking.
Questions developers ask before integrating
Why are snippet, language, image and tone null on every article?
Those fields only exist on the GDELT side of the merge. A search for "OpenAI" over 2 days returned 94 articles with meta.sources_used ["google_news"] and meta.source_errors ["gdelt: throttled (5s window)"], and every snippet, language, tone and source_country in that response was null. The article count was unaffected, the metadata was not. Read meta.sources_used before you build anything on those fields.
Why is url a news.google.com address instead of the publisher's link?
Because Google News RSS publishes its own redirect URLs and the engine returns what the feed gave rather than guessing the destination. The publisher is still identified: source_domain read cnbc.com and techcrunch.com on measured results, and publisher carried the display name, for example "TechCrunch". If you need the real article link, the tech action returns direct URLs, or hand the story to news-article to fetch and extract it.
What does dedup actually merge, and how do I turn it off?
A measured search returned total_before_dedup 101, removed_duplicates 7 and 94 survivors. Clustering is exact URL first, then host plus path, then a stemmed token-set Jaccard comparison of titles at the similarity threshold, which defaults to 0.6 and accepts 0.3 to 1.0 where higher merges less. Survivors carry duplicate_count and also_reported_by. Set dedup false for the raw firehose.
How is this different from news-article?
news-intel is query in, metadata out: it never fetches an article body, so you get titles, URLs, domains and timestamps but no full text. news-article goes the other way, taking one URL and extracting the readable article from it. The normal pipeline is news-intel to discover the story and news-article to read the pages you picked.
How far back can I search?
GDELT indexes roughly the last three months, which is the real ceiling. timespan accepts 1h, 12h, 1d, 3d, 1w, 2w and 1m and defaults to 1d, and passing start or end overrides it entirely, accepting YYYYMMDD, YYYYMMDDHHMMSS or ISO 8601. The freshness block in the response labels the Google News side as near-live in minutes and GDELT as roughly 15 minutes, because of its update windows.
Which country codes should I pass?
search and its relatives take a GDELT FIPS code such as US, UK, FR, IN or BR, while on top_headlines country selects a Google News edition. The engine normalizes across the two: measured top_headlines calls with country GB and country UK returned identical results, led by bbc.co.uk, theguardian.com and independent.co.uk. The sources directory lists its editions in ISO2 form as DE, ES, FR, GB, HK, IN, QA and US.
Why does outlet_count say 2 when both entries are the same domain?
outlet_count is the size of the story cluster, not the number of distinct publishers. A measured "nvidia earnings" story returned outlet_count 2 with both outlets listed as finance.yahoo.com, because that outlet published two variants of the same story and both landed in the cluster. If you need distinct publishers, de-duplicate outlets on source_domain yourself before counting.
What are the ceilings on one call?
max controls how many articles are pulled per source before de-duplication, accepts 1 to 250 and is clamped at 250, which is GDELT's own limit. batch runs up to 10 queries under one shared call and preserves input order; a measured two-query batch returned 190 articles across the two result groups in about four seconds. Large sweeps work better as several batches than as one oversized max.
What is the News Intelligence API?
News Intelligence API is a ReefAPI endpoint group for news intelligence It returns live JSON through POST requests under /news-intel/v1.
Is the News Intelligence API free to try?
Yes. ReefAPI starts with 1,000 free credits, no card required. News Intelligence calls use the same shared credit balance as every other ReefAPI engine.
Do I need a News Intelligence login or account?
No login to News Intelligence is needed for the API response. You call ReefAPI with your x-api-key header, and the playground can run live examples before you create a production key.
How fresh is the News Intelligence data?
The page example is captured from a live search call, and production requests fetch live data through ReefAPI rather than a static sample.
How many credits does the News Intelligence API use?
News Intelligence actions currently cost 1-2 credits per successful call. Failed or blocked calls are free, and all APIs draw from one credit pool.
Can I call News Intelligence from an AI assistant or MCP client?
Yes. Connect ReefAPI once through MCP and your assistant can call news-intel actions with the same key, credit pool and JSON envelope used by normal REST requests.