News Data Scraping Services & News API

We deliver headline, outlet, timestamp, author and any publicly visible summary, and we mark the record as paywalled. We do not extract protected body text, and we do not use subscription credentials, shared logins or paywall circumvention techniques of any kind.

This is a firm rule rather than a setting. A feed containing improperly obtained paywalled content creates liability for you as the recipient, not only for the provider, and that exposure typically surfaces long after ingestion when it is expensive to unwind.

Median under five minutes from publication to delivery on priority sources, which covers wires, major financial press and the outlets you designate as high-priority. Broader long-tail coverage runs on a longer polling cycle, typically 15 to 60 minutes, because polling 80,000 sources at wire frequency is neither economic nor welcome to publishers.

Every record carries both the publisher's stated publication time and our ingestion time, so you can measure our actual latency yourself rather than relying on our claim. We recommend doing exactly that during the pilot.

We resolve organisations to tickers and registry identifiers, and people and locations to typed entities with context. Accuracy is high on well-known entities and degrades on ambiguous names, small private companies and transliterated names in non-Latin scripts — which is true of every entity resolution system, ours included.

Each entity carries a salience score and a resolution confidence. For screening use cases we recommend filtering on both: it removes most false positives at a small cost in recall, and the tradeoff is tunable to your risk appetite.

Yes, on selected sources. Our archive extends multiple years for wires, major financial press and a substantial set of trade publications. Long-tail and regional sources have shallower history, and some have none before we began collecting them.

During scoping we give you exact date ranges per source rather than a headline claim, because a backtest built on a corpus with unacknowledged gaps produces results that look valid and are not. If your strategy needs uniform history across a specific source set, we will tell you plainly whether that exists.

It helps, provided you use both views. For measuring genuine story count, the canonical records are correct. For measuring reach and republication breadth, the duplicate set is what you want — a story picked up by 200 outlets has different value from one carried by three.

We deliver canonical records with the full republishing outlet list attached, so you can compute both story count and coverage breadth from the same feed. Suppressing duplicates entirely would destroy the reach measurement, which is why we attach rather than discard them.

Yes — 40+ languages, including full-text extraction and entity resolution in the major ones. Multilingual coverage is often the strongest argument for a managed feed, because regional developments surface in local-language press well before international outlets pick them up.

Optional machine translation of headline and body is available, delivered alongside the original text rather than replacing it. We keep the original because translation loses nuance that matters, particularly in regulatory and legal reporting.

Yes, and we recommend it strongly. Filters can be set on entities, entity salience thresholds, topics, outlet tier, country, language, sentiment range and keyword patterns, and combined with boolean logic.

Unfiltered global news is around 200,000 articles per four-hour window in our corpus. Almost no use case wants that. Most clients run a tightly filtered stream for alerting plus a broader scheduled batch for analysis, which keeps both cost and noise manageable.

We quote every news monitoring engagement individually, because a real number depends on scope: source count, record volume, refresh frequency and delivery method. Anyone quoting you a price before understanding those four things is guessing.

Publisher count, latency requirement and whether you need full-text or metadata-only delivery are the main drivers. Sub-five-minute latency costs meaningfully more than hourly.

The process is short: one scoping call, a free pilot on your own sources within 48 hours, then a fixed monthly quote. No per-request metering, no overage billing, and field or source additions are handled inside the retainer rather than re-quoted. Request a quote.

Technically, straightforwardly — we deliver deduplicated, clean, provenance-tagged text in Parquet built for exactly that. Legally, it depends on the sources, and you should treat any provider claiming otherwise with caution.

What we do is document the access status and collection methodology of every source in your feed, so your legal team can assess training rights source by source. What we cannot do is grant rights we do not hold. If training is your use case, raise it during scoping — source selection changes materially, and some sources are better excluded from a training corpus than included.