Popularity Is Not Authority

01The random surfer was a popularity model

PageRank has an intuitive definition, given by its authors, and almost nobody quotes it. It does not describe trust, quality or authority. It describes a person clicking at random: given a page, they keep following links, never press back, and eventually get bored and jump somewhere else. What PageRank computes is the chance you find that person on any given page.

That is a popularity measure in the strict sense — a visit probability. The damping factor, the one parameter everybody knows the value of, is defined in the same passage as the probability that the surfer gets bored. Not a trust decay. Boredom.

The same paper is careful about the distinction elsewhere. Listing what it calls external meta information, it names "reputation of the source, update frequency, quality, popularity or usage, and citations" — five separate things, with popularity and citations on opposite ends of the list. The authors were not confusing the two. The industry did that later.

02The same signal, twenty-seven years later

In February 2025 a Google engineer described PageRank to the Department of Justice in one sentence, and it is not the sentence from 1998. PageRank is a signal about distance from a known good source, and it feeds the quality score.

Distance and visit probability are not the same measurement. One asks how likely you are to be stumbled upon; the other asks how far you sit from a fixed set of trusted starting points. The first is symmetric in the graph and rewards volume. The second implies seeds, and rewards proximity — one link from close by can beat many from far away.

What the quality score itself is called is the second surprise. Not authority. The engineer glosses it as trustworthiness, and describes it as largely static across queries and largely a property of the site rather than the page. So the axis that the industry calls authority is, in Google's own words, a trust axis fed by a proximity measure — while the popularity model that gave PageRank its name has quietly moved out.

03Popularity exists — separately, and redacted

If PageRank stopped being the popularity model, something else took the job. In the same exhibit, under Other Signals, there is a line whose name is blacked out and whose parenthesis is not: a popularity signal that uses Chrome data.

That is the whole entry. The name is redacted, the category and the data source are not. Google measures popularity, it does so from the browser rather than from the link graph, and it files it away from PageRank. The leaked ranking corpus has two fields that fit the shape — chromeInTotal for volume and uniqueChromeViews for breadth — but the redaction sits exactly where the connection would be, so whether these are the same thing is unknown. Note also what the entry does not say: unlike PageRank, which is spelled out as an input to the quality score, this signal is listed under Other Signals with no downstream score named at all. If you want to know which letter of E-E-A-T popularity buys, the exhibit does not answer, and the only Google document that does answers narrowly — the rater guidelines admit popularity as evidence of reputation, but only "for some topics, such as humor or recipes" where "less formal expertise is OK". Where expertise matters, popularity is not admissible as reputation at all.

There is a third behavioural measurement, and it is on yet another axis: clicks are the C of the ABC signals, which feed topicality. So user behaviour turns up twice, once in relevance and once in popularity, and neither of those is the trust axis. Three measurements, three places, one word in common usage.

04Where the model was abandoned

The random surfer only works as a popularity model because of one assumption: that every link on a page is equally likely to be followed. That is what makes the arithmetic clean — divide the page's rank by its number of outgoing links, hand out equal shares.

A granted Google patent throws that assumption out. The reasonable surfer weights each link individually, by font size, by position on the page, by whether it sits in a footer or a sidebar, by the commerciality of the anchor text, by whether it points to the same host. Once shares are unequal, the number is no longer the chance of finding a random visitor. It is a weighted estimate of where an interested reader would actually go.

This is the quiet part of the story. The move from popularity to something else did not happen in a public announcement; it happened in a patent that redefines what the arithmetic means, and the name stayed the same throughout.

05What deserves to be called authority

Google has been asked this directly, and the answer has been consistent for years. There is no authority signal. A senior search quality engineer put it plainly in 2017: no single thing is authority, and what exists is a bundle that together is hoped to raise how authoritative the results are.

The same engineer gives the example that shows what the bundle is for. Two articles on the same topic, one in the Wall Street Journal and one on a fly-by-night domain — with no other information, the Wall Street Journal article looks better. Note what that judgement is not based on: not traffic, not link count, not how many people visit either site.

The word survives in the exhibit too, and it is telling how. The engineer says that if competitors saw the logs they would have a notion of authority for a site — in quotation marks, as something a third party would derive from the data, not as something Google stores under that name. Authority is a conclusion, not a field.

So the practical answer to a question this site gets often: popularity is measured, it is measured from the browser, and it is not the thing you are trying to build. What the quality axis wants is proximity to something already trusted. The companion piece to this one argues the same point from the other end — what a link claims when it is set.

06References