What actually makes a page rank

Keywords

1,159

Sitefire's German keyword map, full SERP pulled for each

Ranking pages

10,054

Every page in the top 20, downloaded and read

Head-to-heads

154,919

Page-vs-page comparisons inside the same result list

Signals tested

Authority, topical fit, language, structure, freshness

The setting

Sitefire tracks a map of 1,159 keywords. For every one of them we pulled the live Google result list from DataForSEO, top 20 SERP results.

We then scraped the content of those 10,054 top-ranking pages. For each page we also bought two authority scores: Page Rating (PR), how much link authority that specific page carries, and Domain Rating by Ahrefs (DR), how strong the whole website is.

Everything in this study is a page that already ranks in Google's top 20. That is an important limit, and we come back to it at the end.

What we tested

We compared two pages fighting over the same search.

Take one search

All the pages Google returned for a single keyword, positions 1 to 20.

Pit them against each other

Every possible pairing. Which one ranked higher, and how did the two differ on each of the 13 signals?

Let the pattern emerge

Repeat across 154,919 duels. Signals that consistently pick the winner are the ones that matter.

Because both pages in a duel faced the identical search, everything about the keyword itself (how competitive it is, what the searcher wants, which market) cancels out. What is left is the difference between the two pages, which is exactly what we wanted to isolate.

Does the result hold up?

Two separate questions matter here. The first is "is this real, or did we find noise?" The second is "how much of ranking does it explain?"

Is the finding real?

chi² = 10,073

A significance test against the possibility that our signals do nothing. Above ~22 already means "almost certainly real" for 13 signals. Ours is 450× that threshold. An F-test on the same data agrees: F = 387, where anything above ~2.2 counts as significant.

How much does it explain?

R² = 3.2%

R² is the share of the outcome the model accounts for. 0% = nothing, 100% = perfect. Ours is small, and honestly so: most of what decides a position is invisible to us (click behavior, engagement, internal linking).

Can it pick the winner?

60.1%

Given two pages on the same search, how often we correctly call which ranks higher. 50% = a coin flip. Rises to 70.5% when the two pages genuinely differ, and 74.9% when they are far apart in the ranking.

Are the signals tangled?

VIF < 1.8

Checks whether signals are secretly measuring the same thing. 1 = fully independent, under 5 = healthy, over 10 = broken. All 13 sit under 1.8, so each one carries its own information.

Read together: we are highly confident about which levers move rank and roughly how much, while accepting that we cannot predict where a given page will land. Those are different claims, and only the first one is supported.

Signal strength

The same 13 signals, measured for what separates a top-10 page from the rest, then top 5, top 3, and the fight for position 1. Higher number means the signal matters more. The bars show strength in the race for position 1.

Standardized effect per signal. *** means below a 1-in-1,000 chance of being luck, ** 1-in-100, * 1-in-20. Grey figures are not statistically distinguishable from zero. Based on 17,045 to 80,750 head-to-heads per column.
Signal	Top 1	Top 3	Top 5	Top 10
Decisive
Page authority (PR) link strength of that one page	1.223***	0.972***	0.767***	0.564***
Page language matches the search	0.502***	0.435***	0.384***	0.322***
Meaningful
URL depth (a specific page, not a shallow one)	0.426***	0.316***	0.207***	0.131***
Search words present in the title	0.398***	0.194***	0.139***	0.142***
Site authority (Domain Rating by Ahrefs)	0.279***	0.232***	0.221***	0.224***
Topical fit of the page text	0.244***	0.287***	0.238***	0.177***
Topical fit of the title	0.223**	0.222***	0.167***	0.125***
Only matters in the fight for #1
Article length (longer is worse)	−0.214***	−0.064	−0.037	−0.013
Page displays a date (showing one is worse)	−0.172***	0.016	0.020	0.027
Search words in the headings	0.171**	0.051	0.057*	0.007
Negligible everywhere
Page is the homepage	0.119	0.058	0.064*	0.083**
Age of the page's date	−0.042	−0.090**	−0.013	−0.054*
Exact keyword repetition in the text	0.040	0.051	0.043	0.047*

What helps a page rank

Page authority is the whole game. It more than doubles in importance between qualifying for the top 10 and taking position 1. No other signal comes close.
Serve the language of the search. The second strongest signal overall. In a German market, German content beats every title tactic combined.
Be a specific page, not a shallow one. Deeper, more precise URLs consistently beat broad category pages.
What the page says beats what the title says. Topical fit of the body text outranks topical fit of the title, and it peaks in the fight for the top 3.
Site authority still counts, separately from page authority. A strong site helps even when the individual page is not strong.

What does not

Repeating the keyword. The weakest measured signal in the study. Keyword density is not a strategy.
Stuffing headings. No measurable effect anywhere except a small one in the race for position 1.
Writing longer. Flat everywhere, and actively harmful among pages competing for the top spot. Shorter wins at the top.
Publishing dates. Freshness barely registers, and displaying a date correlates with losing the number one position.
Being the homepage. Almost no advantage once authority is accounted for.

The uncomfortable finding

We tested how well the full 13-signal model predicts which pages make the top 5 on a keyword it has never seen, and compared it to a model using page authority alone.

Random guessing

30.3%

baseline

Page authority alone

44.8%

one single signal

All 13 signals

44.9%

the full model

The content signals add essentially nothing on top of authority. That is not a contradiction of the table above. Those content effects are statistically certain but small, and small effects rarely flip a discrete top-5 decision. Being confident that something helps is a different claim from being able to predict the outcome.

The honest boundary of this research: we can tell you which levers move rank and roughly how much. We cannot tell you which position you will land on.

What this means for content strategy

Every page in this study had already earned its way into the top 20. Whatever it takes to get there, all 10,054 of them had it. That is almost certainly where good content does its work: relevance is the entry ticket, and our data cannot see the pages that never got in.

But once a page is in the room, the ranking order is decided somewhere else. Across every tier we measured, from scraping into the top 10 to claiming position 1, the gap is dominated by authority, and the authority effect gets stronger the higher you aim.

Content gets you into the top 20. Authority is what carries you to number one.