How Newsful seeks truth

Each morning, one Grok search looks for the most-cited news in the world. It reads X and primary sources first, then state-funded and private outlets across regions, in their own languages. A fixed outlet list does not choose the day.

Stories are ranked by how widely they are covered and discussed today: different owners count, copies of one wire story do not, and conversation on X counts too. No region gets a bonus or a penalty. A truly major event that Western media is not reporting still belongs.

The search follows a few rules.

• A real new event or development from the last 36 hours. Background, anniversaries, and opinion pieces are not news.
• One entry per event, and at most two from one ongoing situation, such as one war.
• Stories from the last two editions are skipped unless there is a major new development.
• Each story names its two principal parties, Side A and Side B.

Every story belongs to one desk. A full edition of ten is:

• Politics, 4. Government, elections, diplomacy, conflict, security, crime and justice, migration and rights.
• Business, 2. Business, markets, the economy, and technology.
• Science, 1. Science, health, climate, the environment, and natural disasters.
• Society, 1. Everyday life, education, work, family, food, travel, and human-interest stories.
• Culture, 1. Arts, film, music, books, entertainment, media, and religion.
• Sport, 1.

Within a desk, the most-cited real news wins. If a desk has too little real news that day, its places go to the next most-cited stories from any desk. A desk is never padded with minor or old stories.

The code then removes repeats: a second story about an event already in the edition, and a third story from one ongoing situation. It does not judge content or viewpoint.

Each story gets its own Grok research pass, using web search and X search together. Grok starts from the two sides and the links found in step 1, and works in this order: the named voices on X, state media for each side, private outlets for each side, then primary sources such as statements, documents, and data.

Documents, data, and named primary records beat narrative polish. Grok is told never to invent a source, a quote, or a link, and to search every view before calling it empty.

Every source is a specific article, post, or document, up to 16 per story. Homepages, tag pages, and Wikipedia are not used. Each source carries one of four labels.

• State-funded. Outlets with major government funding, whatever their charter says: BBC, DW, France 24, Al Jazeera, RT, Xinhua, VOA, and similar.
• Independent. Privately owned outlets, including local and specialist press.
• X / on the ground. A named person's post on X.
• Official. Statements, documents, and data. Governments, the UN, WHO, and NGOs are primary sources, not media views.

Grok assigns the label. The site also keeps a list of known state-funded outlets so they are shown the same way every time. On a story page, each source shows its label and, where known, its country.

Side A and Side B describe this story only. An outlet sits on a side when it carries that party's line, not because of its country or a permanent score. Neutral wire copy goes in Key facts, not on a side. One outlet cannot be on both sides.

Five views are always shown, in this order: X / On the ground, State media Side A, State media Side B, Independent Side A, Independent Side B. Each says only what that group reported. They are not blended. If an outlet leaves out a material fact others carry, its view says so.

X / On the ground is named people only: people involved, officials speaking directly, correspondents, experts, and eyewitnesses. Each line says which side the voice supports or comes from. Outlet accounts and anonymous rumor do not count. Follower count does not matter; what they add does. X is the live record, not the verdict.

Most likely true is Grok's own verdict, based on evidence only. It is published as Grok wrote it. Newsful does not edit it.

It may disagree with every outlet and with X. A state outlet can show what its government claims, but cannot alone prove a disputed fact. The verdict does not split the difference, does not lecture, and states awkward supported facts plainly.

Before a story goes live, the code checks only that it has a headline, a summary, and at least two sources. A story that fails is left out, so an edition can have fewer than ten.

The code does not fact-check stories or test links. Weighing the evidence is Grok's job, during research.

A question typed on the home page gets a story written with the same instructions and a fresh web and X search. A follow-up on a story runs a fresh search too: the saved story is context, not proof, and the answer comes first.

Stories are written in English. Italian and Russian versions are translated by Grok when a reader picks that language.