Ranking signals updates under review: the good, the weak and the verdict

Ranking signal claims arrive faster than anyone can check them. This is the shortlist that survives contact with a published source, split into the documented, the weak, and the verdict on each.

What to take away

The good: signals an operator has published

Watch time and average view duration. YouTube names these in its own documentation and has done since 2012. A video that holds viewers past the point where similar videos lose them gets shown to more of them. The signal is real, the operator says so, and the mechanism is checkable on your own channel.

Click-through rate on search results. Google's own guidance treats a result that gets clicked and then abandoned as a weaker answer than one that gets clicked and held. The published version is about usefulness; the measurable version is dwell time. Both point the same way.

Engagement rate weighted by depth. LinkedIn's published guidance on B2B reach describes comments and shares as stronger than reactions. The company states the direction, not the multiplier, which is the honest shape of almost every operator statement.

Originality and duplication penalties. Google's spam policies name scraped and duplicated content as demoted. This one is unusual because the operator publishes the rule and the enforcement is visible in rankings.

Topical relevance of the source. Google's Search Quality Rater Guidelines describe what raters should look for, and the company publishes the document. It is citable about Google's stated intent, not about the weights behind it.

The weak: signals that are inference

Recency. Everyone repeats it. No operator has published a recency weight, and the effect people see is usually explained by the same accounts posting more often.

Posting frequency. The correlation is real and the cause is not. Accounts that post daily also tend to be accounts with staff, budgets and a beat.

Hashtag count. Instagram's own guidance has moved away from hashtag stuffing. The folk version outlived the advice that produced it.

The three-hour window. This one has no source at all. It circulates because it is specific enough to feel like a leak and vague enough to never be disproved.

Follower count as a reach multiplier. Platform documentation consistently describes reach as a function of content performance, not account size. Large accounts get more reach because they start with more chances, not because the system rewards the number.

The verdict

The ranking signals overview sorts these inputs by who produces them, which is the fastest way to see which ones an operator can be held to.

Why the weak ones survive anyway

A ranking system is observable only through what it returns. Given outputs alone, many different mechanisms produce the same visible behavior, and no amount of sampling separates them. This is the ordinary situation of reasoning about a black box: you can build a model that predicts the outputs well and still be wrong about every internal detail.

Three gaps make the outside view weaker than it looks.

You cannot see the counterfactual. The comparison that would settle a causal question, what this same post would have done under a different configuration, is unavailable to anyone who cannot run both.

You cannot see the population. Outside samples are drawn from what already ranked, and the problem with a sample selected on the outcome does not shrink with more data.

You cannot see the change history. When behavior shifts, the outside observer sees the shift and nothing about its cause. A configuration change, a policy change, a shift in who is using the product this month, and a difference in your own material all look identical from here.

How to test a signal on your own property

You cannot inspect the system. You can run a clean comparison on your own accounts, which is what the weak signals on this list have never had.

The same discipline applies to any organization writing about its own methods. Research bodies publish method statements for exactly this reason, and Pew Research Center's account of how it describes its own methods is citable about Pew in a way that no outside description of Pew would be.

Dating a claim so it stays true

An operator's page can change without notice, and often does. Record what the page said and when you read it, so a later reader can tell whether your sentence has come loose from its source.

The life cycle of a ranking claim shows where an undated page ends up: quoted for years after the rule behind it changed.

Common questions

Is a platform's own page a biased source?

It is an interested source, which is not the same problem. Read it as a statement of what the operator wants to be held to in public. That tells you what they will defend, not what the system does.

What about an engineering blog or research paper?

Same category, with an extra caution. A paper often describes a system that was in production well before publication, sometimes a simplified version of it, and the gap is rarely stated. Cite it about the described method, not about today's product.

Can an independent audit substitute for a published signal?

Only within the access it was granted. An audit run against outputs has the same limits as any outside study. An audit with real access can say much more, and then the question becomes what access was given, which is usually the least documented part.

How do I compare platforms if I cannot cite anything specific?

Compare on structural properties instead of internals: what surface it is, what the unit of success is, what the reader arrived wanting. That needs no claim about anyone's code.