About PolicyNews

PolicyNews collects what UK government departments, Parliament, the statistical agencies and the major policy institutes publish, and turns it into one readable, filterable stream.

How it works

1. Collect. Every source publishes an RSS or Atom feed. We read those feeds directly rather than scraping pages, so we get the publisher's own title, link and blurb.
2. Summarise. A language model writes a short, factual summary of each item from its title and the publisher's blurb, tags it with up to three topics from a fixed list, and rates how much a policy professional would want to see it.
3. Rank. The default view blends that rating with recency, so a significant report from yesterday outranks a routine notice from this morning. Switch to Newest first for a pure chronological feed.

Government research

The research view separates commissioned research — evaluations, survey work, qualitative studies and impact assessments — from the general run of government news, and attributes each report to the agency that carried out the work.

That attribution is looked up, not guessed. The agency is almost never in a report's title, but GOV.UK indexes the full document text, so we search it directly for each agency by name. Two caveats follow from that: a report is tagged if it mentions an agency, so an evidence review citing several firms will be tagged with all of them (about 1.5% of reports carry four or more tags for this reason), and an agency we do not yet track will show no tag at all rather than a wrong one.

What the score means

The number on each item is an editorial interest score from 0 to 100. It reflects substance, not the prestige of the publisher: a routine procedural notice from a major department scores low, a rigorous finding from a small institute scores high. It is a machine judgement and it will sometimes be wrong — treat it as a skim aid, not a verdict.

Honest limitations

• Summaries are generated automatically and are not checked by a human. Always read the original before citing anything.
• Every headline links straight to the publisher. We reproduce only a short summary and never the full text.
• Coverage is limited to organisations that publish a machine-readable feed and allow automated readers. A few notable institutes block us at their edge; they are listed on the sources page.
• Publication dates come from the feed. A handful of publishers date items loosely, so ordering can be approximate.

Using the data

There is a JSON endpoint at /api/articles and an RSS feed at /feed.xml. Both accept the same q, kind, topic, source, days, sort and limit parameters as the site.

Currently tracking 5,498 articles from 37 live sources. Last updated 3h ago.