The short version
AI gathers the information and writes the page. A second AI, which did not write it, checks each claim against the sources and corrects or removes what does not hold up. No person reviews a page before it is published. Every page lists the sources it relies on.
Step by step
- Research. An AI researcher looks up the topic on the live web. It is told not to write from memory, because what an AI learned in training goes out of date. It may only rely on pages it actually opened, and only on sources that meet our sourcing standards.
- Writing. The same AI writes the page in its own words, ties each number, date and named event to a source, and lists those sources. Statements a company makes about itself are presented as that company's claims.
- Independent fact-check. A second AI receives the finished page with one job: try to break it. It opens the cited sources, checks every claim it can identify, and then edits the page. Claims the source contradicts are corrected. Claims that overstate the source are toned down. Claims it cannot support are removed. Weak or dead sources are replaced with better ones, or the claim goes.
- Automated checks. Software confirms that every link in the text also appears in the source list, that no source is Wikipedia, and that source links load.
- Publishing. The page goes live with the date it was last updated.
What this cannot guarantee
The process reduces errors. It does not eliminate them, and you should know where it is weak.
- AI makes mistakes. It can misread a source, miss context or state something more firmly than the evidence allows. A second AI catches many of these, not all.
- Sources can be wrong. We choose credible ones, but a credible source can still publish an error or be overtaken by events.
- Things change. A page is accurate as of its last-updated date at best. Fast-moving subjects go stale quickly, which is why time-sensitive statements carry dates.
- Some sources resist checking. Certain sites block automated readers. Those pages are read another way where possible, but a few are harder to confirm than others.
So treat every page as a guide to its sources. For anything that matters to you, follow the links and read the original.
Keeping pages current
Trackers are on a schedule. Fast-moving ones are checked for news every day and the rest every week. The process is the same for all of them. An AI looks for new developments, and a second AI checks them against their sources. Only then is the tracker updated. Each tracker shows when it was last checked and a history of what changed.
New articles are written most days, with the news categories taking turns. Each one goes through the steps above. When there is no real development to report in a category, the article explains a subject instead of dressing up old news as new.
An article's research also feeds the rest of the site. An event it confirms can be added to a tracker or to the timeline of an organization or country profile. A term it relies on can get its own glossary entry. An organization at the center of the story can get a profile. Each of those changes is checked against its source in the same way before it is published. Glossary entries and profiles are not re-checked on a fixed schedule of their own; each shows the date it last changed, so you can judge how current it is.
Where a page mentions a term, organization or country that has its own page here, the first mention is linked to it automatically. When a check finds an error, it is fixed and logged as described on Corrections.
Last updated