How we work

At a glance

• Every story is compared by the Lumen AI engine, our own system for grouping and comparing news coverage.
• The engine writes only from what the linked outlets reported. It adds no facts, names, numbers or quotes of its own.
• Each story names the outlets behind it and links to every original report.
• A story publishes on its own only when it is at least 400 words and built on at least 5 rated outlets from at least 2 sides.
• We never say which side is right. We show what each side reported and what it left out.

How a story is made

1. 1Collect. We read the public feeds of news outlets across the political spectrum, every 30 minutes.
2. 2Group. Reports about the same event are grouped together by comparing what each headline and summary says, within a 48-hour window.
3. 3Rate. Each outlet carries a lean rating (left, centre or right). We count how many rated outlets on each side covered the event.
4. 4Draft. The Lumen AI engine writes one account of the event from those reports, then compares how each side covered it: what each led with, what one side reported and the other left out, and where they disagree.
5. 5Check. The draft must clear the quality bar below. Drafts that fall short wait for an editor instead of publishing.
6. 6Link. Every story lists every outlet report it was built from, with a link to the original.

The Lumen AI engine

The Lumen AI engine is the software that turns many outlets' reports into one comparison. It uses a large language model to draft the account and the side-by-side comparison, under written rules that every draft must follow:

• Use only facts present in the outlets' reports, and prefer facts more than one outlet reported.
• Attribute any claim only one outlet reports, and say so when outlets disagree.
• Neutral wording: no loaded adjectives, no guesses about motives, no conclusions.
• Describe what each side emphasised without judging whether it is right.
• Answer the questions readers ask about the event only where the reports can answer them.

Every story page says it was compared by the engine and shows how many outlets on each side it drew on. If you see a sentence that the linked reports do not support, that is an error and we want to hear about it.

How outlets are rated

Each outlet carries a lean rating based on published outlet ratings and our own review of its coverage. Outlets without a rating are read but not counted on any side. The rating describes the outlet, never a single report.

Coverage labels

• Balanced: the shares of left- and right-leaning outlets are within 12 points of each other.
• More from the left / right: one side covered the event noticeably more than the other.
• Blindspot: fewer than 15% of the rated outlets covering the event are on one side.

The quality bar

A story publishes automatically and is offered to search engines only when it meets all three conditions:

• At least 400 words of reporting and comparison.
• At least 5 rated outlets covered the event.
• At least 2 of the three sides covered it.

Drafts that miss the bar wait for an editor. Shorter stories published before this standard remain readable, but we have asked search engines not to index them.

How much we publish

We publish a limited number of stories a day and raise that number only as the quality holds. Since 1 October 2026 the limit starts at 12 stories a day and steps up every week or two, reaching 80 a day after week 8. The biggest events, covered by the most outlets, go first.

Corrections and contact

When a story gets a fact wrong, we correct it on the page and update its modified date. To report an error, a missing outlet or a rating you disagree with, write to privacy@thelumen.io with the story link.