Why a Source Is the Whole Claim
A fact and its source are one object. Pull them apart and what is left looks identical to what you started with, which is precisely the problem: a sourced claim and an invented one are the same shape on a screen. Typography is free. So is a confident tone, a domain name ending in .org, and a logo.
Here is what that buys. In the Stanford study of how experienced internet users evaluate what they read, participants were given five minutes on a site called minimumwage.com, which publishes research-styled arguments against raising the minimum wage. Its own About page described it as a project of the Employment Policies Institute, “a nonprofit research organization” sponsoring “nonpartisan research”. Both are the work of a Washington public relations firm that lobbies for the restaurant and hotel industries, whose speciality, in the New York Times’ description, is creating “official-sounding nonprofit groups to disseminate information on behalf of corporate clients”. None of that appears anywhere on the site. When the five minutes were up, 80% of the Stanford undergraduates had spent no time at all finding out who was behind it. They had spent the five minutes reading the arguments, which is what the site was built to have them do.
What Counts as a Source
Not everything shaped like a citation is one.
- A link to the record. The filing, the ruling, the register, the dataset, the photograph, the transcript. This is the good case: you can go and look, and what you find either says the thing or it does not.
- A link to reporting that saw the record. Second best, and most of what anyone can offer. It transfers the question from “is this true” to “does this outlet check”, which is a question you can answer once and reuse.
- A link to an article making the same claim. Often worthless, and the most common thing you will be handed. Three outlets citing each other in a circle is one unsourced claim wearing a crowd.
- A name with no link. “Experts say”, “studies show”, “sources confirm”. Every one of those has a real version, and the real version names who.
The test is not whether the writer sounds careful. It is whether the writer left you a way to disagree with them.
The Question Underneath a Number
Numbers are where unsourced claims survive longest, because a number reads as a measurement even when nobody measured anything. When you meet one, the useful question is not whether it is plausible. It is: who counted, what did they count, and over what period?
The ICIJ’s own Offshore Leaks Database is a good demonstration of a source being careful about itself. It publishes more than 810,000 offshore entities from five leak investigations, and it says on the way in that “many of the activities carried out through offshore entities are perfectly legal”. A name in that database is a fact about a corporate structure. It is not a finding about a person, and reading it as one would be inventing a claim the source explicitly refuses to make.
That distinction is the whole of this beat. Almost every case we cover turns on the gap between what was documented and what was proven, and a reader who collapses the two ends up believing things nobody established.
What We Do about It
The rule this desk runs on is that nothing enters our system without a source URL, and nothing publishes without passing the checker that enforces it. That is not a promise about our character. It is a build step: a claim with no link attached fails and does not render. The five rules we bind ourselves to are all like that, deliberately—an aspiration is something you can quietly stop meeting.
You do not have to take that on faith either. Open any story here and follow three links. If they do not say what we said they say, tell us.