Everyone is citing the same mistake
Ten citations can look like ten pieces of evidence even when they all return to one uncertain claim.
Written Aug 2026 · Updated Sept 2026
I see a claim in five respectable articles.
Each article has a citation beside it.
At first this looks like five pieces of evidence. Then I open the citations.
Four articles point to the fifth. The fifth points to a paper that said something narrower. Sometimes that paper points somewhere else too.
There were never five pieces of evidence.
There was one uncertain claim and a lot of links.
A citation is not evidence#
A citation tells me where an author says a claim came from.
It does not tell me that the cited paper contains the claim. It does not tell me the method was sound. It does not tell me the new sentence preserved the old one accurately.
But citations look like verification. The number creates confidence before I have followed any of them.
This is the mistake: counting references as if they were independent witnesses.
Sometimes they are all repeating the same witness.
The claim changes as it travels#
A paper might say:
In this small sample, the result suggests X may be associated with Y.
A review shortens it:
X is associated with Y.
A later article writes:
Studies show that X causes Y.
Then a presentation says:
Scientists have proved Y.
Nobody needs to invent a complete falsehood. Each person only removes one caveat because it feels inconvenient, already understood or too difficult to explain.
By the end, the confident sentence may have more citations than the cautious one it replaced.
Many visible citations can still lead back to one weak or misread source.
This has been measured#
Steven Greenberg studied a network of medical papers around a claim about the protein beta amyloid in a muscle disease. He traced every statement and citation in the literature he examined.
He found citation bias, amplification and invention. Papers that supported the claim were cited more, while papers that did not support it were isolated. As the network grew, review articles helped turn the claim into apparent authority.
The interesting part is that the authority was produced by the network itself. The number of papers increased without a matching increase in direct evidence.
Ole Bjørn Rekdal documented another version of the problem through stories about iron in spinach. The popular explanation blamed a misplaced decimal point. That explanation was itself poorly sourced and repeatedly copied. The correction had become another legend.
Even scepticism can inherit the same bad habit.
Why the mistake survives#
Following every citation is slow.
Academic papers are difficult to access. Reviews are easier to read than the studies underneath them. A claim may sit outside the author's main field, so a secondary source feels safer. Deadlines reward finishing the paragraph, not reconstructing the citation family tree.
There is also a status effect. Once enough respected people repeat something, checking it can feel unnecessary or even arrogant.
Search engines make this stranger. Highly cited work becomes easier to find. New authors cite what they can find. The work becomes more visible because it is cited, then more cited because it is visible.
The loop can reward agreement before accuracy.
Count roots, not links#
When a claim matters, I want to know how many independent roots it has.
My process is:
- Open the source attached to the sentence.
- Search the paper for the exact subject of the claim.
- Check whether the paper reports original evidence or cites another source.
- Preserve the strength of the language: may, associated with and causes are not interchangeable.
- Look for evidence that disagrees, not only another paper that repeats the claim.
- If several articles lead to one study, count one study.
This does not mean primary research is automatically true. It means I have finally reached the thing that can be evaluated.
Write the uncertainty down#
The cleanest protection is to leave a trail.
If a number comes from a secondary source, say so. If the original paper is not available, say that. If a result only appeared in one small study, do not hide that because the sentence reads better without it.
The uncertainty belongs to the claim.
Removing it is not editing. It is changing the information.
The question I now ask#
When everyone seems to agree, I do not first ask how many people said it.
I ask whether they all learned it from each other.
Ten links may be useful. But if they all return to one root, I still have one piece of evidence.
And sometimes, after following the root, I have none.