How fact-checking works
Someone reads the draft looking only for claims, not for style. Every statement that could be true or false gets marked: figures, dates, names, job titles, quotations, prices, product features, legal and tax statements, anything phrased as “research shows”. Each marked claim is then traced to a source, and the source is recorded next to it. Anything that cannot be traced is rewritten so it no longer asserts a fact, or removed.
The quality of the source decides the value of the check. A regulator’s own site, a company’s own documentation, a published dataset or a named person you actually spoke to are primary. Another marketing blog repeating the claim is not a source; it is the same claim wearing different clothes. Where a number has travelled from article to article for years without anyone naming its origin, the honest move is to drop it rather than join the chain.
Why fact-checking matters
A single wrong claim spends trust you took a long time to build, and it is usually the reader who knows the subject best who spots it — the person you most wanted to impress. On money, health, legal and immigration topics the stakes are higher again, because a wrong statement can cause a reader real harm. Those are the pages search engines hold to the strictest standard, described as YMYL.
Checking also has a quieter benefit. Doing it forces you to notice how many sentences in a normal marketing draft are asserting things nobody has verified, and how much stronger the piece reads once those are either sourced or cut.
Common mistakes with fact-checking
The biggest is trusting an AI draft. Language models produce fluent, confident sentences and will supply a citation that looks correct and does not exist. Everything in AI-generated content that reads like a fact needs checking, including the sources it offers, and that check takes longer than people expect when they plan the workload.
The second is checking the interesting claims and skipping the boring ones — an out-of-date regulation, a renamed government department, a product feature that moved to a different plan. The third is borrowing a foreign figure for a local market. Nepal-specific data is often thin or old, and a benchmark from another country dressed up as local is simply a wrong claim. Say the local figure is not available and describe the mechanism instead.
How to act on it
Make the writer supply sources with the draft rather than leaving the checker to find them; a claim nobody can source is cheaper to cut at drafting than after publishing. Keep the sources somewhere durable — a comment thread, a footnote file, a column in the content sheet — because you will need them the day someone challenges the piece.
Set a rule for what happens to unverifiable claims, and apply it without argument: rephrase as an opinion, attribute it to whoever said it, or delete it. Re-check anything time-sensitive when you refresh the page, since a verified fact goes stale quietly. And treat the whole thing as part of editorial review rather than an optional extra pass that gets dropped whenever the deadline tightens.
