How It Works

The same event can look like a very different story depending on where in the world you read it. The Intelligence Bulletin exists to make those differences visible — and it does the work through an autonomous, agentic editorial system that monitors the global press, decides what matters, gathers the world’s coverage, and writes each bulletin. The system is governed by a written editorial charter: a standing definition of what deserves an edition, what the coverage must support, and when publishing nothing is the right answer. Here is the whole process, start to finish.

1. We monitor the world’s news, continuously

Around the clock, we scan thousands of news articles from across the global press, drawing on GDELT — the Global Database of Events, Language, and Tone — which monitors print, broadcast, and online news in over 100 languages. That wide, multilingual net is how we spot which stories the world is actually paying attention to, rather than seeing the news through the narrow window of any single country’s media — including stories that may be dominating coverage abroad while going almost unmentioned at home.

2. We build a shortlist of the day’s most significant stories

From that firehose, the system distils a shortlist of candidate stories, each carrying signals about its importance: how widely and prominently it is being covered, how many distinct outlets and countries are involved, the key people and organisations at its centre, and where in the world it is unfolding.

Two of those signals deserve mention, because most aggregators do not compute them. We measure each story’s coverage breadth — how many countries’ own press genuinely originated reporting on it, rather than republishing someone else’s — and its wire saturation: how much of the apparent coverage is a single agency dispatch appearing under many mastheads. Thirty sites carrying the same wire story are one voice, not thirty, and the system knows the difference before an editorial decision is ever made. Breadth here is earned by measurement, not by quota: because genuinely multi-sourced reporting outranks wire amplification, the shortlist reaches past the loudest story of a season on merit.

3. An editorial AI chooses what to publish next — against a written standard

An editorial agent weighs those candidates the way a newsroom editor would — but without a personal agenda, and against an explicit test. It favours stories of genuine global significance, and especially those involving multiple countries across different blocs, because those are the stories where perspectives truly diverge and a comparison is worth making. Under the charter, every pick must clear two bars: the story must genuinely matter — not merely be widely syndicated, because ceremonial and procedural events flood the wires while leaving readers no wiser — and the way different countries cover it must be non-obvious: a divergence a reader could not have predicted from the headline. An unexpected split is worth your time; so is an unexpected agreement between rivals. A purely domestic story, however large, often yields only one viewpoint; we would rather bring you a story the world is actively arguing about. No single running story is allowed to quietly become the only thing we cover: the system remembers what we have recently published, so an ongoing story is revisited only when there is a genuinely new development — never simply restated — and it has to earn the slot against everything else available that day.

4. We gather a deliberately diverse set of sources — and verify who they are

Once a story is chosen, the system seeks out coverage of that exact event from a wide range of outlets across borders and blocs — the countries at the centre of the story, the country where it is unfolding, and observers with no direct stake in it — because distance from an event changes what an outlet is willing to say about it. The goal is not balance for its own sake, but range: enough genuinely different vantage points that the contrasts become visible.

Before any outlet is presented as a country’s voice, we verify that it actually is one. A website named after a country with no newsroom in that country is not treated as that nation’s press. A broadcasting service funded by one government but aimed at another country’s audience is attributed to its actual publisher, not its target. An independent newsroom operating in exile is identified as exactly that, and an article credited to a wire agency is treated as the agency’s voice, whatever masthead it appears under. When a nation’s own press is silent on a story, we say so — silence can be part of the story — rather than dressing up a substitute.

5. We separate fact from framing

From that body of coverage we summarise the verified core of the story — the facts most sources agree on — and then set out how the framing differs: which details each side emphasises, which language it chooses, and what it leaves out. Fact and interpretation are kept clearly distinct. Claims about what an outlet omitted are made only when we have verified the omission against its actual reporting; an absence we have not confirmed by reading is not reported as one.

6. We publish — and cite everything

The system runs every few hours, around the clock — but a bulletin is published only when a story earns it. Every draft passes a final review against the charter’s standard before it goes out; drafts that fail are held, and the slot stays empty. If nothing has genuinely moved since the last edition, we would rather publish nothing than repeat ourselves. On quiet days the slot simply stays empty rather than being filled with a comparison the coverage does not support. Every one links back to the original reporting it draws on, so you can always reach the primary sources and judge for yourself. Each edition also states how many independent reporting chains its sources represent, and labels relayed, translated or wire-derived coverage on the source card itself — so you can see how much genuinely separate reporting stands behind a story. We summarise and point you onward; we do not republish other outlets’ work in full, and all source material remains the property of its respective publishers. Bulletins are also shared to our newsletter and to LinkedIn, X, and Instagram.

The role of AI — and its limits

We are open about the fact that The Intelligence Bulletin is AI-generated. The system is designed to be accurate, well-sourced, and even-handed, and removing individual editors from story selection is a deliberate choice to reduce bias. Accountability, however, stays human: every bulletin is published under a named author, and responsibility for what appears here rests with a person, not a model. The written charter is part of that accountability: the system’s behaviour follows the charter, and changing the behaviour means changing the charter first, on the record, by a person. We also select and configure the underlying AI models with even-handedness in mind — part of the same effort to keep bias out of what you read. But automation is not infallibility: an AI can misread nuance or make a mistake. That is why every bulletin is traceable to its sources, and why we take corrections seriously: when we get something wrong we say so on the bulletin itself and record it in a permanent, public corrections log — we do not silently edit. If you spot an error, please tell us — see our Editorial Standards and Contact page. The Intelligence Bulletin is published by LevTech Global and is not affiliated with Google, GDELT, or any news organisation.