How Does Google Actually Rank Pages? Every Ranking System Explained

"The algorithm" is a convenient shorthand and a misleading one. Google doesn't rank pages with a single algorithm: it runs every query through a pipeline of dozens of specialized systems, each doing one job, stacked and layered together. Google's own documentation names 17 of them publicly. Two more, arguably the two that matter most for how a page actually moves in the results, aren't in that documentation at all: one surfaced only because a federal antitrust trial forced Google's own VP of Search to describe it under oath, the other is a patent Google filed in 2018 and has never confirmed using in live ranking.

This is a full accounting of both sets: what Google will tell you directly, and what's only known because of litigation and patent filings. Every claim below is sourced, primary where a primary source exists.

What Google Actually Documents: 17 Named Systems

Google publishes a page called "How Google's Search ranking systems work," and it's the closest thing to an official parts list the company has ever released. It names 17 systems it currently runs, each with a narrow, specific job. None of them is "the algorithm"; together, applied to every query, they are.

Google is explicit that this list is representative, not exhaustive. The real number of signals and sub-systems feeding into ranking is far larger and mostly undisclosed, for the same reason Google has never published a full ranking-factor list: naming a specific lever invites people to optimize directly for the lever instead of for genuine quality.

The Four That Retired Into Core Ranking

Four systems that used to run standalone no longer exist as separate systems. Each was folded into Google's core ranking systems, meaning its function still runs, just no longer as an isolated, individually-toggleable system:

• Panda (2011) targeted thin and duplicate content; integrated into core ranking in 2015. Full history in History of Google Panda.
• Penguin (2012) targeted link spam and manipulative link schemes; integrated into core ranking in 2016 and made real-time. Full history in History of Google Penguin.
• Hummingbird (2013) rebuilt query understanding around meaning rather than keyword matching; evolved into today's core language-understanding systems.
• Helpful Content System (2022) demoted content written primarily to rank rather than to help people; merged into core ranking in March 2024, covered in the Helpful Content Update Survival Guide.

For the complete chronology of every core update and spam update these systems intersect with, see the Google Algorithm Updates Timeline.

The System Not on That List: NavBoost

For years, Google gave carefully worded public answers about whether clicks directly influence rankings. That ended in the 2023–2024 United States v. Google LLC federal antitrust trial, when Google's own VP of Search, Pandu Nayak, testified under oath and described NavBoost as "one of the important signals that we have." It has never appeared on Google's public ranking-systems page. Everything reliably known about it comes from trial testimony and exhibits, not a Google blog post.

What the trial record describes: NavBoost is a re-ranking system, applied after an initial candidate set of pages is retrieved, that scores results using historical click behavior aggregated across a rolling 13-month window (an 18-month window before 2017). It distinguishes between goodClicks (the user was satisfied and didn't return to the results page), badClicks (the user "pogo-sticked" back quickly, a signal of dissatisfaction), and lastLongestClicks (the final result a user settled on and stayed with). Trial exhibits also confirmed a separate Chrome-derived popularity signal, with metrics like chrome_trans_clicks and uniqueChromeViews feeding directly into the same systems, drawing on Chrome's roughly two-thirds share of the browser market.

The scale is the part that matters strategically, not just technically: testimony indicated that Google's 13-month click dataset is roughly equivalent to over 17 years of data available to its nearest search competitor. That gap is why NavBoost is described in court filings as closer to a "memorization system," it doesn't reason about a page's quality from first principles, it remembers, at massive scale, which results actually satisfied past users for a given query, and boosts accordingly.

Information Gain: The Real Patent Behind the Buzzword

"Information gain" gets used loosely across SEO content in 2026, often as a synonym for "add original data to your article." The actual source is more specific and narrower than that usage suggests: US Patent 11,354,342, "Contextual estimation of link information gain," filed by Google on October 18, 2018 and granted June 7, 2022, invented by Victor Carbune and Pedro Gonnet Anders.

What it actually describes: a machine-learning system built to solve a specific redundancy problem, mainly framed around voice assistants and conversational search, where a user might be told the same troubleshooting steps or facts across multiple documents in a row. The system builds a semantic representation of documents a user has already seen, compares it against candidate documents on the same topic, and outputs an information-gain score reflecting how much genuinely new content a candidate document adds. Higher-scoring, less redundant documents get prioritized for presentation.

Two things are true at once here, and conflating them is where most of the SEO content on this topic goes wrong. First: the patent is real, specific, and describes a genuine, sound approach to reducing redundant results. Second: Google has never confirmed this patent is an active production ranking factor in Search, let alone confirmed the widely repeated 2026 claim that a specific core update made it "the dominant content-quality signal." That claim traces to SEO industry blogs analyzing post-update volatility, not to any Google statement. It's a reasonable hypothesis given the patent exists and matches observed patterns, it is not a confirmed fact, and this article isn't going to present it as one.

How the Pieces Actually Fit Together

Read individually, a list of 17-plus systems looks like an arbitrary pile of acronyms. Read as a pipeline, the design logic is straightforward, and it explains why some optimization advice works and other advice is superstition:

Stage 4, highlighted, is the one system in this pipeline Google has never named in its own public documentation.

1. Retrieval. Crawling and indexing systems build the searchable set of documents; this stage decides what's even eligible to be considered, before any quality judgment happens.
2. Relevance and understanding. RankBrain, BERT, neural matching, and passage ranking work out what a query and a candidate page actually mean, beyond literal keyword overlap, and produce an initial relevance-ranked set.
3. Quality and trust. Reliable information systems, original content systems, the reviews system, and core ranking (which now includes the former Panda, Penguin, Hummingbird, and Helpful Content functions) reweight that set based on quality and trustworthiness signals.
4. Behavioral re-ranking. NavBoost applies historical click satisfaction data on top of the quality-ranked set, the layer that isn't in Google's public documentation but that trial testimony confirms runs late in the pipeline, after relevance and quality have already been assessed.
5. Final filtering. Deduplication, site diversity, spam detection, and removal-based demotion systems clean up the result set immediately before it's shown: no duplicate entries, no single domain hogging the page, no confirmed spam or policy violations.

The practical implication: a page can be relevant and well-written and still lose at the behavioral-reranking stage if it doesn't actually satisfy the people who click it, and a page can satisfy everyone who clicks it and still never get the chance if it fails the quality or spam stage first. Optimization advice that only targets one stage, keyword-matching for relevance, backlinks for the old PageRank-only model, "add more words" for quality, tends to underperform because it ignores every other stage the page also has to clear.

What This Means If You're Optimizing in 2026

None of this changes the honest answer to "what's the one thing to fix": there isn't one, by design. But the pipeline view does clarify where effort actually compounds. Genuinely original reporting or data (original content systems, and plausibly information gain if the patent is live) survives the quality stage. Content that resolves the query well enough that people stop searching (satisfied clicks, lastLongestClicks) survives the NavBoost stage. Neither substitutes for the other. A page that's technically correct but generic can pass the relevance stage and still lose the behavioral stage to a less polished page that actually answers the question faster.

Credify's Google Penalty Risk Scanner checks a URL against current spam-policy and quality signals directly, useful for the stages Google does document. The stages Google doesn't document, NavBoost's click-satisfaction behavior specifically, aren't independently checkable by any outside tool, including this one; the closest available proxy is Search Console's own click-through and position data for your actual queries.

Frequently Asked Questions

How does Google actually rank pages?

Google ranks pages by running a query through dozens of specialized systems rather than one single algorithm: retrieval systems find candidate pages, relevance and quality systems like RankBrain, BERT, and the reliable information systems score and re-order them, click-based systems like NavBoost re-rank based on historical user satisfaction, and filtering systems like deduplication, site diversity, and spam detection remove or demote pages before the final results are shown. Google's own documentation names 17 of these systems publicly; others, like NavBoost, have only become known through litigation.

How many ranking systems does Google actually use?

Google's official "How Search ranking systems work" documentation names 17 active systems by name, plus four more (Panda, Penguin, Hummingbird, and the Helpful Content System) that were folded into core ranking and retired as standalone systems. Google has stated the real number of signals and sub-systems is far larger and not fully disclosed; the named list is a representative, not exhaustive, account.

What is NavBoost and why isn’t it in Google’s official documentation?

NavBoost is a click-based re-ranking system that Google's VP of Search, Pandu Nayak, described under oath as "one of the important signals" Google uses, during the 2023–2024 United States v. Google LLC antitrust trial. It scores results using a rolling 13-month window of aggregated Chrome and Search click data (goodClicks, badClicks, and lastLongestClicks), then re-ranks an initial candidate set based on which results actually satisfied past users. It has never appeared in Google's public ranking-systems documentation; everything known about it publicly comes from court testimony and trial exhibits, not a Google blog post or help page.

What is Google’s Information Gain patent?

Information Gain refers to US Patent 11,354,342, "Contextual estimation of link information gain," filed by Google in October 2018 and granted in June 2022. It describes a machine-learning system that scores how much new information a document offers beyond what a user has already read on the same topic, then prioritizes higher-scoring, less redundant documents. Google has not confirmed this patent as an active ranking factor in Search; the widely repeated claim that a 2026 core update made it "the dominant ranking signal" originates from SEO industry commentary, not from Google.

What happened to Panda, Penguin, and Hummingbird?

All three were retired as standalone systems and folded into Google's core ranking systems. Panda (2011, thin/duplicate content) was integrated into core in 2015. Penguin (2012, link spam) was integrated in 2016 and became real-time. Hummingbird (2013, query understanding) evolved into current core language-understanding systems. The Helpful Content System (2022) followed the same path, merging into core ranking in March 2024. None of the four exist as separately named systems today; their functions are part of the broad core ranking systems that run continuously.

Related Reading

For the full chronology of every named update these systems trace back to, see the Google Algorithm Updates Timeline. For what replaced Panda, Penguin, and the Helpful Content System specifically, see History of Google Panda, History of Google Penguin, and the Helpful Content Update Survival Guide. For how the quality bar these systems enforce is defined day to day, see What Is E-E-A-T?. To check your own site against current spam-policy and quality signals, use the free Google Penalty Risk Scanner or the E-E-A-T Checker.

Primary sources: Google Search Central: How Search Ranking Systems Work · US Patent 11,354,342: Contextual Estimation of Link Information Gain · United States v. Google LLC (U.S. Department of Justice case page)