How the brief is made

Every issue is produced by software, from the first step to the last. This page says what that software does, so you can judge how far to rely on it.

The sources

Once a day a program collects new articles from a fixed list of feeds: UAE dailies, regional outlets, the shipping and energy trade press, and a Russian-language news search. It keeps what concerns the Gulf and drops what it has offered before. Every item in the brief names the outlet it came from and links to that page.

Three passes, all by AI

The curator. An AI model (Claude, made by Anthropic) reads the day’s candidates and drafts the brief in English. It keeps an item only if it changes something for a UAE resident, and it writes from the source in its own words.

The editor. A second pass starts fresh, sees only the draft, opens the page behind every item and checks each number, date, name and claim against it. It hedges what the source hedges and cuts what it cannot confirm. It may cut an item or a whole issue. It may not add a fact.

The translator. A third pass renders the edited English into Russian.

The gate

Before anything is published, a program that is not an AI model checks the format, that every link came from the day’s feeds and still opens, and that the Russian carries the same lines and the same numbers as the English. An issue that fails is not published that day.

What no person does

No person reads an issue before it goes out. The publisher set the rules and wrote the checks; the daily decision of what runs, and how it is worded, is made by the software described here.

An AI checking an AI can still be wrong. It can misread a source, and it can miss what a careful human editor would catch. That is why every item links to its source, and why corrections are published in the open.

What it never does

It does not reproduce a source’s text or images: it summarises and links. It does not publish unattributed reports of strikes. It does not say what to buy, sell or hold.

The weekly quiz

Once a week an AI model writes five to seven questions from that week’s published items. A program checks that each question names its item and that the right answer is a number or a name that item states. What is shown under an answer is the item’s own first line, not new text. No person reviews a quiz either.

Since when

English issues begin on 5 October 2026, and so does the separate editor pass. Earlier issues, from 29 July 2026, exist in Russian only and were made by an earlier version of this process.