Google Ranking Signals

I have stopped arguing with people about ranking signals on the basis of "Google said so." Not because Google lies constantly, but because public statements are marketing-reviewed simplifications, and we now have two better windows into the machine: sworn testimony from United States v. Google, and the Content Warehouse API documentation that leaked in May 2024. Neither window gives you weights or formulas, but together they killed several polite fictions the industry had been fed for years.

Bucket one: what Google confirms

The boring confirmed stuff still does most of the work. Google's ranking systems documentation openly describes signals built on content relevance and helpfulness, link-based authority (PageRank is cited by name), freshness for queries that deserve it, and page experience factors. HTTPS was announced as a lightweight signal back in August 2014 on the Google Search Central blog. Spam systems like SpamBrain sit on top, demoting or removing what the other systems would otherwise reward. None of this is controversial, and honestly, if your site fails on these fundamentals, the exotic signals below are irrelevant to you.

Bucket two: what the DOJ trial forced into the open

The antitrust trial testimony was the single most educational SEO document of the decade, and it cost us nothing. Google's Pandu Nayak testified about Navboost, a system that memorizes user click behavior on queries, trimmed to roughly 13 months of data. Trial exhibits described how Google distinguishes clicks, hovers, scrolls, and what internal slides called the difference between a click and a satisfied click. In other words: aggregated user interaction data feeds ranking. The industry spent fifteen years being told clicks were too noisy to use, while a system was quietly using them at scale. That one still annoys me.

Testimony and exhibits also referenced deep-learning rankers like RankBrain, RankEmbed and DeepRank operating alongside the classic signals, which fits the public picture of ranking as a pipeline: retrieve candidates with core signals, re-rank with machine learning, adjust with user data.

Bucket three: the 2024 Content Warehouse leak

In May 2024, internal Google API documentation covering roughly 2,596 modules and over 14,000 attributes surfaced publicly, first reported by Rand Fishkin of SparkToro and analyzed in depth by Mike King of iPullRank. The documents describe attributes like siteAuthority (a site-level authority score, after years of Google saying it uses nothing like domain authority), Chrome-related view data, and flags around content freshness, authorship, and demotion classes.

Here is the caveat everyone skips: the leak documents storage schemas, not the live ranking function. An attribute existing in a data warehouse does not tell you whether it is used in scoring today, at what weight, or whether it was deprecated in 2019. Treat leak attributes as hypotheses worth testing, not confirmed signals. Anyone selling you a "leak-optimized SEO package" is selling confidence, not knowledge.

Signal reference table

What to actually do with all this

The practical takeaway from Navboost is that the click is part of the product. Your title and snippet earn the click, and your page has to end the search. In Search Console, hunt for queries where you rank in positions 4 to 10 with CTR well below the position average: those are snippet problems, and snippet problems are now ranking problems. Then look at your highest-traffic pages in analytics for quick returns to the SERP; that pattern is exactly what a satisfied-click system would punish over time.

The takeaway from the leak is site-level reputation is real. Thin sections drag on strong sections. Prune or consolidate content that nobody searches for, no one links to, and Google will not index, because it sits on the same site-level ledger as your best work.

FAQ

Source: https://dixonjones.com/google-ranking-signals/