Fact Checking Policy

What we check, and when

Fact checking at NEWS ANALYSIS happens before publication, not as an occasional exercise afterwards. Every article is checked against its sources as part of writing it, and the checks below are the minimum rather than the exception.

The checks we run on every article

Does the source say what we say it says. Each factual claim is traced back to the document, filing, study, transcript or data set it came from. Where our sentence paraphrases, we confirm the paraphrase preserves the meaning, including any conditions the original attached to the claim.

Are the numbers right. Figures are matched against the source for value, unit, currency and time period. Percentage changes are checked to confirm the baseline. Totals are checked to confirm they are totals and not subsets. Where a source gives a range or a confidence interval, we do not report the most striking end of it as the finding.

Are the names and titles right. People, organisations, products, places and job titles are verified against a current source. Titles change, companies are renamed, and products are revised, so a detail that was accurate a year ago may not be accurate now.

Are the dates right. We check when something happened, when it was announced, and when it takes effect, because those are frequently three different dates and conflating them changes the story.

Is the attribution correct. We confirm that a quoted or summarised position belongs to the person or organisation we attribute it to, and that it is represented in context rather than trimmed into a different meaning.

How we treat single source claims

A claim that rests on one source is reported as such, with the source named, unless we can corroborate it independently. We do not turn a single unverified assertion into apparent fact by attributing it to reporting in general. Where several outlets carry the same claim, we check whether they are independent or whether they all originate from the same release or briefing, because repetition is not corroboration.

Studies and research

When we cover research, we read the paper rather than the press release about the paper. We check the sample size, what was actually measured, whether the finding is correlational or causal, and whether the authors themselves described limits that the coverage has dropped. A study of a small group under controlled conditions is not evidence of a general effect, and we try not to write as though it were.

We note when research is preliminary, not peer reviewed, or funded by a party with an interest in the result. None of those facts automatically invalidates a finding, but a reader is entitled to know them.

Claims about the future

Forecasts, targets, projections and roadmaps are reported as intentions or estimates, never as events. We name who is forecasting, and where possible we note the track record of similar forecasts. A company saying it will ship a product on a date is a statement of intent, and we write it that way.

What we do when we cannot verify something

We leave it out, or we publish it with the uncertainty stated plainly. An interesting claim we cannot stand up is not improved by hedging language that implies we can. If a central element of a story fails verification, the story does not run in that form.

After publication

Checking does not stop at publication. When a reader challenges a fact, we go back to the original source rather than defending the sentence as written. When new information changes the picture, we update the article and record the change. Our Corrections page explains how those changes are labelled.

If you believe we have published something inaccurate, our Contact page describes how to tell us and what we will do with the report. Specific challenges with a pointer to the underlying source are the most useful thing you can send us.