A claim becomes interesting to us in two places. The first is when we see it once. The second is when we see it twice. The first occurrence is information. The second is a question, and the question is whether the second source saw it independently or copied the first.
The instinct most people have when a fact shows up in two articles is to treat it as more credible. We have learned to do the opposite. We treat repetition as a prompt to walk backwards through the citation chain, source by source, until we find where the claim actually originated. The chain is shorter than people assume, and what we find at the end of it usually changes how much weight we give the claim.
What the chain usually looks like
A typical chain has the shape of a pyramid pointed downward. At the top, there are five or six articles, all of which mention the fact. Some cite each other. Some cite no one. The ones that cite no one usually link to a different article in the same group, which cites another article in the same group, which finally links out to a single primary source. Sometimes that primary source is a press release. Sometimes it is a paper. Sometimes it is a single transcript of a podcast.
The first surprise, when we started doing this systematically, was how rarely the primary source said what the secondaries said. The first secondary article usually adds a small simplification: a percentage gets rounded, a hedged statement gets a confident reframing, a future tense becomes past. The second secondary picks up the simplified version and compresses again. By the time the claim reaches the fifth or sixth article, it has lost its original conditions and become a flat assertion. The chain looks long, but the original is one source, and it is saying something narrower than the chain reports.
The three endings we find
In our experience, the chain usually terminates in one of three ways, and each ending demands a different response.
The first ending is the one we hope for. We find a primary source, the secondaries are roughly faithful to it, and the claim survives the trip. The primary source might still be wrong, but at least we are evaluating one piece of evidence rather than the misleading appearance of five.
The second ending is the lossy one. We find a primary source, but the secondaries have drifted from it in ways that matter. The primary says “in some configurations.” The secondaries say “always.” The primary says “we observed.” The secondaries say “this is true.” When we find this, we write down the primary source’s actual wording and discard the secondaries. The secondaries were not lying, exactly. They were compressing, and compression is where claims pick up the confidence they did not start with.
The third ending is the one we expected to be rare and turned out to find often. We follow the chain, and it ends nowhere. Every article cites another, every citation eventually loops back into the same set of articles, and no one in the chain links to an original observation. Sometimes we can identify the article that started the loop. Sometimes we cannot. The claim has the shape of a fact but no origin. When this happens, we record it as unsourced and either drop it or qualify it. We do not pretend that five articles citing each other is the same as five observations.
What this is not
This practice can look like skepticism for its own sake. It is not. We are not trying to disprove claims. We are trying to count them honestly. The question we are answering is not “is this true” but “how many independent observations support this.” Five sources citing each other is one observation. One primary source with five faithful summaries is also one observation, and we should treat it that way when we decide how much to bet on the claim.
The other thing this is not is exhaustive. We do not trace every claim back to its origin. Most claims do not warrant the effort. The ones we trace are the ones we are about to write down ourselves, or the ones a downstream decision will rest on. The rest we accept conditionally and flag in our notes for later, if they become load-bearing.
Why the chain is shorter than it feels
The reason this works at all is that the public web is less voluminous than it appears. A topic that has hundreds of articles often has three or four original sources behind it. The rest is reweaving. We used to find this depressing. We have come to find it useful. A small primary corpus is searchable. A vast secondary corpus is not.
There is also a practical signal in how a source handles its own citations. A primary source tends to describe its method: where the data came from, what the measurement was, what the conditions were. A secondary source tends to describe its conclusion. When we open a page and find ourselves reading a conclusion without a method, we have learned to look further down the chain. It does not always mean the source is unreliable. It does mean we are not yet at the end of the chain.
The work this protects us from
Most of the bad claims we have caught ourselves about to publish were not invented. They were inherited from a chain we had not walked. The chain had taken a hedged statement and stripped it of conditions across four hops, and we had picked it up at hop five thinking we were reading a fact. The trace is what catches this, and the trace only takes a few minutes if the chain is short.
It also catches something subtler: the false sense of corroboration. When we have walked a chain and found that all five articles trace back to the same press release, we cannot use the five as independent evidence for anything. They are one source, and the press release has its own incentives. That is a useful thing to know before we cite it.
We are not always patient enough to do this. Some weeks we publish things we wish we had traced. But over time, the trace has changed which claims we trust on contact. A piece of information that surfaces in multiple places now produces a different feeling than it used to: a sharper interest, a slightly raised eyebrow, and a small mental note to find out where it came from before we lean on it.