How our news pages are made

Several times a day we search the news for a fixed set of plain terms (places, people and institutions, never a loaded phrase) and read the feeds of publishers we follow. Links about the same event are grouped into one story. We link to the original article and never host its text.

A story's source count is the number of different outlets covering it, however many articles each one ran.

Each outlet is described in two separate ways: its political lean, in the usual American sense (left, center, right), and how it covers Israel (staunchly pro-Israel, supportive, critical but fair, or hostile). They are separate because they are different questions. A left-leaning outlet can be staunchly pro-Israel.

A rating is first proposed by AI, from search results about the outlet and from the outlet's own recent headlines, and is marked provisional. An editor then confirms or corrects it. A rating an editor has confirmed is never changed by AI afterwards. Criticism of Israeli government policy is not hostility: most mainstream outlets are rated critical but fair.

Some outlets cover Israel in a way we judge hostile: state media of hostile regimes, outlets that routinely present terror groups' claims as fact, or that deny Israel's right to exist. We do not send our readers to them.

We still count them. How widely a story is covered is a fact about the story, and leaving them out would make the numbers wrong. So a story may say “23 sources · 4 not linked”: all 23 are in the count and in the coverage bars, and 19 are listed. We do not name the outlets we do not link.

The bars on a story page are counted by our software from each outlet's rating. They are never written by AI. Every bar adds up to the story's source count, and outlets nobody has rated yet are shown as “not rated” rather than guessed.

“Underreported” marks a story several pro-Israel outlets are carrying while the mainstream press is barely covering it.

A summary is written by AI from the headlines and descriptions of every article in the story and the full text of a few of them, chosen to span different kinds of outlet. It is in our own words, and the articles are read and not kept.

Every sentence shows the sources it rests on. Our software checks each one: a sentence that cites an article that is not part of the story is thrown away before you see it. When a sentence also rests on an outlet we do not link, it says so.

“Where coverage differs” appears only when there are at least two groups of two outlets to compare; otherwise we leave it out rather than fill the space. It describes how outlets framed a story and does not accuse them. An editor can correct a summary, and an edited summary is labelled “AI summary, edited”.

Under a story we show posts, videos and articles from organizations and people listed in our directory that are about that same event. We show disagreement between them when there is any. We never label a person or organization as left or right.

If a rating, a summary or a grouping looks wrong to you, tell us. The report box under the news feed reaches an editor.