Why one page beats another

For almost any query, the index holds not the ten pages you see but potentially millions that contain the words. Ranking is the job of ordering that haystack so the ten most useful sit on top.

No single rule does it. A modern engine blends a large number of signals, and the exact recipe is secret and constantly changing. But the signals group into a small number of questions, and thinking in questions rather than in a list of tricks is what separates understanding search from chasing it.

• Relevance. Does this page match what the query is about?
• Authority. Is this a source worth trusting?
• Quality. Is the page itself good, useful, honest, well-made?
• Context. Who is asking, from where, on what device, meaning what?

Everything marketed as a "ranking factor" is a proxy for one of those four. Get the questions and the specific tactics stop being a mysterious list and become obvious consequences.

The oldest signal is relevance: does the page concern what the query is about? Early engines answered it crudely, by matching keywords, which is why the early web filled with pages stuffing a phrase in repeatedly. Counting words is trivial to game.

Modern relevance works on meaning, not string matching. The engine represents both the query and the page as points in a semantic space, the embeddings idea from machine learning, so a page about "how to fix a slow laptop" can rank for "my computer is lagging" despite sharing almost no words.

This is why keyword density stopped being a useful lever. The engine is not counting a phrase; it is asking whether the page is about the thing the query means. Two consequences follow:

• Writing naturally and covering a topic thoroughly beats repeating a target phrase.
• Search intent matters more than wording. "Best running shoes" wants a comparison; "buy Nike Pegasus" wants a shop; "running shoe" wants a definition. Same topic, three different pages should win, because the intent differs.

Matching the intent behind the words is most of relevance now.

Relevance alone is not enough, because thousands of pages are relevant and many are junk. The idea that made Google was to judge authority from the web's own structure: PageRank.

The core insight is elegant. Treat a link from page A to page B as a vote by A for B. A page with many links pointing to it is probably important. But not all votes count equally: a link from an already-important page is worth more than one from an obscure one. So importance is defined recursively, a page is important if important pages link to it, and the math resolves that circular definition across the whole web graph.

The deeper why: links were a signal that was expensive to fake. Anyone can write that their page is authoritative; getting genuinely respected sites to link to you was hard, so links approximated real-world trust the words on a page never could.

That original elegance is also why the next step in the story is an arms race. The moment links became the currency of ranking, an entire industry formed to manufacture them.

Once links decided rankings and rankings decided money, links became something to buy, trade, and fabricate: link farms, paid links, comment spam, entire networks of sites existing only to link to each other. The signal that worked because it was expensive to fake was now being faked at scale.

Search engines responded, and the back-and-forth shaped modern SEO:

• Weighting links by the trust of the linking site, so a thousand junk links count for little.
• Ignoring or penalising obvious manufactured-link patterns.
• A tag that lets a site link without passing a vote, so citing a page does not endorse it, which matters for user-generated content and paid placements.

The general dynamic here is worth naming, because it recurs everywhere in ranking: any signal that predicts quality becomes a target the moment it is rewarded, and then degrades as a predictor. It is Goodhart's law again, the same trap that turns engagement metrics into clickbait. Every ranking signal lives on a treadmill: useful, then gamed, then defended, then partly retired.

This is why there is no fixed list of what works. The list is a moving battlefield.

Links work poorly for a lot of the web. A brand-new but excellent page has none yet. Whole topics, especially health, finance and other high-stakes areas, need a judgement of trustworthiness that raw link counts do not capture.

So engines added quality signals aimed at the content itself. Google publishes a public framework for this, summarised as E-E-A-T: Experience, Expertise, Authoritativeness, Trustworthiness. It is not a number in the algorithm; it is the concept the algorithm's quality signals try to approximate, and the standard human raters use when they evaluate results to train it.

The practical translations are concrete: is there a real, identifiable author with genuine standing on this topic; does the page show first-hand experience rather than rewording others; is the site transparent about who runs it; are claims sourced. This is why an anonymous, sourceless article struggles on a serious topic no matter how well optimised.

The recurring theme: as each mechanical signal gets gamed, engines lean harder on signals that are harder to fake, and genuine expertise and transparency are among the hardest of all.

Before any page is scored, the engine interprets the query, and the same words can mean different things.

• Location. "Coffee" from a phone means coffee near me now; the best answer is a map, not an encyclopedia entry.
• Time. "Election results" wants the most recent, so some queries trigger a strong freshness preference and others do not.
• Device and history shape it further; "python" leans one way for someone who reads programming sites and another for a nature reader.
• Ambiguity. "Jaguar" is a car, an animal, and a team, so the engine often deliberately diversifies the results to cover the likely meanings rather than betting on one.

Two people can search identical words and correctly get different, personalised results, which is one reason "what rank am I" has no single answer.

The reframing that matters: an engine is not matching a query to pages. It is matching an interpreted intent, in a context, to pages. Relevance and quality are scored against that interpretation, so understanding what a query is really asking is upstream of everything else.

The way ranking is usually taught, a checklist of two hundred factors, is close to useless, because the factors change constantly, most are unknown or secret, and their weights shift with every update. Optimising a list you cannot see is a fool's errand.

The durable model is the four questions, and every real tactic is downstream of one of them.

Notice what every row rewards: being genuinely the best answer. That is not an accident or a platitude. An engine's own incentive is to satisfy searchers, so over time, and unevenly, it keeps adjusting toward signals that track real quality and away from ones that can be faked.

The honest strategy that survives every algorithm update is therefore the least clever one: be the best answer to a real question, and make it easy to crawl, understand, and trust. Everything else is a temporary proxy for that.

The query is interpreted into an intent and context first, candidates are pulled from the index, then scored against relevance, authority and quality, and finally ordered, often diversified across meanings.