Google’s Ranking Factors, Evidence-Graded: What the Leak, the Trial, and a Decade of Studies Actually Prove

The recycled-listicle problem

Search “Google ranking factors” and you’ll get roughly the same list, republished every January with a new year in the title: title tags, backlinks, content length, Core Web Vitals, sprinkle in “E-E-A-T,” done. Most of it traces back to correlation studies from the 2010s, restated as if they were confirmed mechanics. Some of it is outright folklore Google has explicitly denied for a decade.

Two events in 2024 changed how much of this we can actually verify. First, in March 2024, more than 2,500 pages of internal documentation from Google’s Content Warehouse API — describing 2,596 modules and roughly 14,000 internal attributes — were published to a public GitHub repository, apparently by mistake, and later analyzed publicly by SparkToro’s Rand Fishkin and iPullRank’s Mike King. Second, sworn testimony from Google’s own Vice President of Search, Pandu Nayak, in the Department of Justice’s antitrust case against Google (United States v. Google LLC) put some of Google’s ranking mechanics on the public record under oath — including several statements that directly contradicted years of public messaging from Google spokespeople.

Neither of those sources hands you a weighted formula. Google has been consistent, before and after both events, that raw attribute names in leaked documentation don’t tell you what’s actually used in live ranking, or how heavily. That caution is fair — and it’s exactly why this piece grades every claim by where it actually comes from, instead of treating all of it as equally solid.

The Evidence Tier framework

Before ranking any factor, ask one question: who said so, and how would they know? Every claim in this piece — and, frankly, every claim in every ranking-factors article you’ll ever read — falls into one of four tiers.

Tier 1: what Google has actually confirmed

This is the shortest list, and that’s the point — it’s short because it’s solid.

• Tier 1 Core Web Vitals, updated. In March 2024, Google officially replaced First Input Delay (FID) with Interaction to Next Paint (INP) as the third Core Web Vital, alongside Largest Contentful Paint and Cumulative Layout Shift. This was announced and documented directly by Google, not inferred.
• Tier 1 The Helpful Content System. Google’s people-first content guidance, originally a standalone classifier introduced in 2022, was folded into Google’s core ranking systems in 2023–2024. Google’s own documentation says it rewards content demonstrating first-hand experience and penalizes content produced primarily to manipulate rankings.
• Tier 1 Scaled content abuse is an explicit spam policy. Google’s spam policies name the practice of publishing large volumes of content — AI-generated or otherwise — primarily to manipulate search rankings, independent of whether any individual page is well written.
• Tier 1 Mobile-first indexing is complete. Google confirmed it now uses the mobile version of a page’s content for indexing and ranking for essentially the entire index.

Notice what’s missing: no confirmed weighting, no confirmed “you need X backlinks” threshold, no confirmed word count. Google has repeatedly said publicly that it does not use a fixed checklist — and on this narrow point, Tier 1 evidence and Google’s messaging actually agree.

Tier 2: sworn testimony and the leak — where the real news is

This is the tier that actually moved the conversation in 2024, because it’s the first time outside observers got anything resembling primary-source material instead of inference.

The leak, briefly

In March 2024, internal documentation for Google’s Content Warehouse API — reportedly pushed live by an automated internal tool — sat briefly on a public GitHub repository before being discovered. Erfan Azimi of EA Eagle Digital found it, corroborated it with former Google employees, and shared it with Rand Fishkin, co-founder of SparkToro, on May 5, 2024. Fishkin spent weeks verifying it before publishing jointly with Mike King of iPullRank on May 27, 2024. The documentation describes 2,596 modules and roughly 14,000 attributes — module and field names, not a ranked list of factors, and critically, not their weights.

Google’s response, delivered through Search Liaison Danny Sullivan, was that outside parties risked drawing incorrect conclusions from documentation lacking the context of how — or whether — each attribute is actually used in live ranking. That’s a legitimate methodological caveat, and it applies to every specific attribute name floating around SEO Twitter since: a field existing in an API schema is not proof it drives rankings today.

The DOJ testimony, briefly

Separately, in the 2023 proceedings of United States v. Google LLC, the federal antitrust case that a U.S. District Court ultimately ruled against Google in August 2024, Google’s own Vice President of Search, Pandu Nayak, testified under oath about ranking mechanics — testimony that, unlike the leak, Google cannot dismiss as decontextualized, because Google’s own executive said it.

The headline revelation was NavBoost, a click-based re-ranking system Nayak testified has existed since roughly 2005 — directly contradicting years of public statements from Google representatives (including former head of web spam Matt Cutts) that click-through data was too noisy and too easy to manipulate to use in ranking.

How a page actually gets ranked, per the testimony

The most useful thing to come out of the trial isn’t any single factor — it’s the shape of the pipeline. Ranking is staged, not a flat scorecard:

Two things are worth sitting with. First, NavBoost isn’t a fine-tuning tweak — testimony describes it acting early, culling tens of thousands of candidates down to a few hundred before the more expensive machine-learned systems ever see them. If your page doesn’t survive that click-based cull, later-stage quality signals may never get the chance to matter. Second, “twiddlers” — Google’s internal term for post-hoc adjustment algorithms — exist specifically to correct cases where click data alone would produce bad results, like an engaging-but-inappropriate page outranking a duller, more authoritative one. NavBoost is powerful, but testimony indicates it does not have unchecked authority over the final result.

Tier 3: correlation studies — useful, but read the fine print

Correlation studies from SEO analytics vendors are the backbone of most “ranking factors” content, and they’re not worthless — they just answer a narrower question than most articles imply. A correlation study can tell you that pages ranking on page one tend to share a trait. It cannot tell you that adopting the trait causes the ranking, because top-ranking pages might simply share some third cause — bigger budgets, older domains, more resources for original research — that produces both the trait and the ranking.

The honest way to use Tier 3 evidence: treat it as a signal about what good pages tend to look like, not a spec sheet for what to build. Write the length the topic actually requires. Earn links by being citation-worthy, not by hitting a target number.

Tier 4: folklore and denied claims still floating around

These persist in briefs and client calls years after Google addressed them directly:

• Tier 4 The meta keywords tag affects rankings. Google confirmed this was ignored well over a decade ago; it still shows up in “on-page checklists” today.
• Tier 4 A fixed “sandbox” penalty holds back all new domains for a set period. Google representatives have given inconsistent public answers here; treat it as unconfirmed rather than as a rule to plan around.
• Tier 4 Domain age, by itself, is a ranking factor. No Tier 1 or Tier 2 source confirms domain age as an independent signal; older domains simply tend to have more of the things that are associated with rankings — content, links, history.

None of this means these tactics never “worked” for anyone. It means the mechanism claimed for why they worked doesn’t hold up against the sources we now actually have.

A mistake I made, and an unpopular take

A self-audit checklist

Before publishing or refreshing a page, run it against what’s actually verified:

• Does the title tag and opening section make the page’s value obvious enough to earn a click and survive the moment right after the click — the “lastLongestClicks” concept from the leaked schema, in plain terms?
• Have you checked Interaction to Next Paint alongside Largest Contentful Paint and Cumulative Layout Shift — the current, Tier 1 Core Web Vitals — rather than optimizing for the retired First Input Delay metric?
• Does the page demonstrate first-hand experience or original analysis, per Google’s own Helpful Content guidance, or could it have been assembled entirely from other people’s pages?
• Is the word count driven by what the topic needs, not a target pulled from a correlation study?
• Would this page still be worth publishing if no one ever linked to it — i.e., does it stand on its own value, independent of the Tier 3 signals it might eventually attract?
• Have you removed any tactic on this page whose only justification is Tier 4 folklore?

Frequently asked questions

Sources

• Rand Fishkin, “An Anonymous Source Shared Thousands of Leaked Google Search API Documents With Me,” SparkToro, May 2024 — sparktoro.com
• Barry Schwartz, “Google Search Document Leak Reveals Ranking Details,” Search Engine Land, May 2024 — searchengineland.com
• Hobo Web, “Key Strategic SEO Insights from the U.S. DOJ v. Google Antitrust Trial,” analysis of Pandu Nayak testimony — hobo-web.co.uk
• JC Chouinard, index of DOJ trial exhibits including the Testimony of Pandu Nayak — jcchouinard.com
• Google Search Central, Core Web Vitals documentation (INP replacing FID, March 2024) — developer.chrome.com
• Google Search Central, Helpful Content and Spam Policies documentation — developers.google.com

This piece synthesizes public reporting on the 2024 Content Warehouse API leak, publicly available DOJ trial materials, and named third-party correlation studies. Where a claim rests on secondary analysis rather than a primary document we could verify directly, it is graded Tier 2 or Tier 3 rather than presented as confirmed.