Reviews
Part of A ranking signal is not the same thing as a ranking factor
How a claim about ranking signals turns into folklore
Ranking signals timeline of a claim, from one account's odd week to folklore: seven stages, why aggregation is the dangerous one, and how to date a claim.
Claims about ranking behave like rumors and follow a rumor's life cycle. Someone sees a number move, says why, and the explanation travels further than the observation, losing conditions along the way.
It returns a year later as a rule everyone knows; this page tracks that process stage by stage, with the check that belongs at each stage.
What to take away
- The evidence in a ranking claim is at its strongest on day one and decays from there, while the claim's confidence moves in the opposite direction.
- Most claims never had a falsifiable form. They were compatible with any outcome from the moment they were written, which is why nothing ever kills them.
- A claim with no date attached is not old or new. It is unreadable, and the right response is to discard it rather than to test it.
Seven stages
| Stage | What exists | What has been lost | The check that still works |
|---|---|---|---|
| Anomaly | One account sees one number move | Nothing yet, but there is also no comparison | Was anything else changed that week |
| Anecdote | A public post naming a cause | The other candidate causes | Ask what the poster expected before they looked |
| Aggregation | Several people say the same thing | The independence of the reports | Did they read each other first |
| Article | A written explanation with a mechanism | The conditions and the sample | Which sentence would be false if the claim were wrong |
| Advice | An instruction to do something | The word "sometimes" | What is the cost if the instruction is wrong |
| Folklore | A rule stated without a source | The origin | Try to trace it back two steps |
| Contradiction | A visible counterexample | Nothing, but nobody updates | Watch whether the rule is retired or narrowed |
The stages are not a slide from truth into error. Stage one is often correct and always underdetermined. The damage is done between aggregation and advice, where the qualifiers fall off and a description of one account's week becomes a general instruction. That week is also one surface, and the same material behaves differently on every surface.
Seven stages of a claim
- Anomalyone account sees one number move
- Anecdotepublic post naming a cause
- Aggregationseveral people say the same thing
- Articlewritten explanation with a mechanism
- Adviceinstruction to do something
- Folklorerule stated without a source
- Contradictionvisible counterexample, nobody updates
Why aggregation is the dangerous step, not the last one
The intuition is that many people saying the same thing is stronger evidence than one person saying it. That holds only if the reports are independent.
In practice the second report was written by somebody who read the first, went to look at their own numbers with a hypothesis already in hand, and found what they were looking for. That is not confirmation. It is circular reporting with a delay of a few days, and it produces exactly the pattern that reads as corroboration.
The same shape appears when a belief spreads because it is being repeated rather than because it is being tested. An availability cascade does not require anyone to lie. It only requires that repeating what you already heard is cheaper than checking it, which it always is.
Dating a claim, and why it matters more than judging it
Two claims can be identical in wording and have opposite value, and the only thing separating them is when they were made and what the system looked like then.
Three dates to check
- Date the observation was made
- Date it was written up
- Date it was last reviewed
So the first question about any piece of ranking advice is not whether it is true. It is when it was written, against what, and by someone with access to what. The overview explains why a stable list of ranking signals is not on offer, so this decay is not a fixable defect in the writing.
Three dates matter, and they are usually different:
Dating a claim
- The date the observation was made.
- The date it was written up.
- The date it was last reviewed by anyone.
Most advice you will find carries none of the three. A page that has been quietly edited for five years shows only the newest date, which describes the edit and not the claim. Treating an undated claim as current is the single most common way to act on something that stopped being true before you read it.
What survives the cycle
Some things do not go stale, and they are the only part of the corpus worth keeping. Claims about constraints survive: a system with more items than slots must select, whatever it is doing this quarter.
Claims about categories survive: knowing that an input comes from the audience rather than from you tells you something durable about your control over it. Claims about method survive: how to compare two periods honestly does not depend on the system.
What does not survive is anything with a number, a weight, an order of importance, or a named mechanism. Those claims get repeated most, because they are the most useful if true. They are the first to expire.
The stage map of a ranking pipeline is durable in its shape and undated in its details on purpose.
Reading the cycle backward
Given a claim, you can usually place it: folklore has no author, advice has an author and no conditions, and an article buries conditions in a paragraph most readers skip.
An anecdote has a person and a week.
Walk a claim back two steps toward its anomaly, and you learn more than arguing about whether it is right.
When the walk back ends at a single account's week, you have learned the real strength of the evidence, and it is usually enough to justify a test and nothing more. Somebody else's aggregate will not settle it for you.
Common questions
Is there any point in reading ranking folklore at all?
Yes, as a source of hypotheses. Folklore accumulates around places where people notice something, and the noticing is often real even when the explanation is not. Read it for where to look, never for what to conclude. Testing what it points at is a separate job, and how to run your own audit sets out that design.
How long is a specific ranking claim good for?
Nobody can tell you that, including the people who publish the claims, because the half-life depends on how often the underlying system changes and that is not disclosed. Treat any specific claim as unsupported once you cannot establish when it was made.
Why do contradicted claims keep circulating?
Because retirement requires someone to do work with no reward attached. The person who first published a rule gains nothing by withdrawing it, and the people repeating it were never invested enough to check. Nothing in the cycle removes a claim, so the corpus only grows.
Does the same cycle apply to what platforms publish themselves?
The first stages do not, because a platform's own document is a primary statement rather than an inference. The later ones absolutely do: a paraphrase of a paraphrase of an official page reaches folklore just as fast, and the general problem of reading a feed's behavior back to a cause is where most of those paraphrases go wrong.
