This page explains how we choose issues, where our numbers come from, how we calculate them, how articles are checked, how we use AI, and how we correct mistakes.
Where do issues and numbers come from?
Issues start from what people in each language are searching for and reading right now: search trends and business news headlines in each of our seven languages. Then we apply one test: does this move someone’s money? We cover an issue only if it does, through jobs, prices, pensions, taxes, housing, rates, investments or a company’s fortunes. Pure sports results and celebrity news are left out unless money is at the center of the story.
For numbers, primary sources come first. Company figures come from regulatory filings and the companies’ own releases, such as SEC EDGAR in the US and DART in Korea. Economic figures come from official statistics, and market prices and index levels from market data providers. News headlines are used only to find issues and see why something is in the news; we link to them and never copy their text.
Every figure in an article is tied to a source and an as-of date in our data package. An automated check blocks publication if a number cannot be traced to that package.
Which numbers do we calculate ourselves?
When an issue is about a company or a market, the numbers that make an article worth reading are usually our own calculations, and we label them as such:
- Quarterly figures derived from cumulative filings, and growth rates.
- Valuation measures: market value (price times shares outstanding), trailing P/E (market value over the last four quarters of net income) and P/E on company guidance (price over guided EPS times four).
- Earnings quality checks: operating cash flow against net income, receivable days and inventory growth.
- Historical patterns: what happened four quarters after a company’s past margin peaks, or over twenty trading days after similar price moves.
- An estimated window for the next earnings report: the next quarter end plus the company’s past reporting delay. It is always labeled as an estimate.
How are articles reviewed?
Every article goes through the same steps before publication:
- An interview: for every article, Vizer answers a few short questions about the issue. The views in the article come from those answers.
- A number check: an automated check matches every number against its sourced data and blocks publication if one does not match.
- A claim check: every statement that is not a number is marked as confirmed by an official source, reported by the press, the author’s own view, or our inference. We search for a source before calling anything unsupported.
- A cold read: a reviewer who knows nothing about how the article was made reads it the way a first-time reader would, title included.
- No buy or sell calls: the checks reject buy or sell advice, and when an article touches something Vizer holds, the article says so.
Vizer reviews each article after publication and corrects it where needed.
How do we use AI?
AI tools help draft articles, adapt them from the English original into the other six languages, and carry out the claim check and the cold read. Data collection, calculations and the number check are done by our own code, so every number in every language edition matches the same data package. The views in an article are Vizer’s, taken from the interview, not generated by AI.
How do we correct mistakes?
We check the report against the source and update the article. Significant corrections are noted at the top of the article. We update existing articles rather than publishing duplicates on the same topic.
Our standards
We write analysis, not recommendations. Outlooks are framed as conditions and scenarios. Reader polls show reader opinion, not facts or forecasts, and we compare the results across our seven language editions.