SOURCE

A guide to Google Search ranking systems

Properties1
tagsseo, google, official, ranking-systems, pagerank, helpful-content
AUTHOR
Google Search Central
PUBLISHED
2025-12-10
ORIGINAL
view original

5 min read

A guide to Google Search ranking systems

Author: Google | Published: 2025-12-10 | Source: article

Summary

Google’s ranking-systems guide is the official inventory of notable systems that score hundreds of billions of pages. Two framing rules matter more than any named system. Ranking is designed to work at page level, using many signals to rank individual pages. Site-wide signals and classifiers also exist and contribute to understanding pages — but good site-wide signals do not mean all content from a site will rank highly, and poor site-wide signals do not mean every page will rank poorly. Systems are tested continuously; Google publishes notices when updates are useful to creators.

Named current systems include: BERT (word combinations and intent); crisis-information systems (personal-crisis hotlines; SOS Alerts); deduplication (hide near-duplicates, including not repeating a featured-snippet URL later on page one); an exact-match domain system that down-weights domains built to match a query (“best-places-to-eat-lunch”); freshness / query-deserves-freshness; link analysis including PageRank, still “part of our core ranking systems” and evolved since the original paper; local news systems; MUM (not used for general ranking; specific applications); neural matching; original-content systems (original reporting ahead of citers; canonical markup support); removal-based demotion (volume of valid legal or personal-information removals can demote a site); passage ranking (score sections of a page); RankBrain (words related to concepts; can rank pages that lack the exact query terms); reliable-information systems (authoritative pages, quality journalism, content advisories); a reviews system for insightful, original, expert/enthusiast reviews; site diversity (generally ≤ two listings from one site in the top results; subdomains usually treated as the root, sometimes split); spam detection including SpamBrain.

Retired / absorbed: the Helpful content system (2022 “Helpful Content Update”) became part of core ranking systems in March 2024; Hummingbird (2013); Panda (2011, into core 2015); Penguin (2012 link-spam, into core 2016). The list is not a ranking recipe and is not complete; it is the official named-systems map. The two framing rules — page-level scoring, and site-wide signals that do not automatically lift every URL — are the claims this wiki should keep even when a new named system appears.

Key Claims

  • Page-level ranking. Systems rank individual pages; site-wide signals help interpret pages but do not lift every URL. (official)
  • PageRank is still a core ranking system, among many link-analysis systems; it has evolved since 1998. (official)
  • Exact-match domains are discounted, not boosted. (official)
  • Passage ranking can use a section of a page for relevance. (official)
  • RankBrain / BERT / neural matching match meaning, not just exact keywords. (official)
  • Site diversity generally caps a domain at two top listings. (official)
  • Helpful content system is retired as a standalone — folded into core in March 2024. (official)
  • Panda and Penguin are historical names now inside core systems. (official)
  • Dedup, freshness, original-content, reviews, spam (SpamBrain), and legal-removal demotions are separate named systems. (official)

Notable quotes

“Our ranking systems are designed to work on the page level, using a variety of signals and systems to understand how to rank individual pages. Site-wide signals and classifiers are also used and contribute to our understanding of pages. Having some good site-wide signals does not mean that all content from a site will always rank highly, just as having some poor site-wide signals does not mean all the content from a site will rank poorly.”

Why citable: Official kill-shot on “DA/DR/site authority lifts every page” folklore. Also the right frame for leak attributes like site-level quality scores: stored ≠ applied to every URL.

“Among these is PageRank, one of our core ranking systems used when Google first launched. … How PageRank works has evolved a lot since then, and it continues to be part of our core ranking systems.”

Why citable: Links are not dead. Official: PageRank is one core system among many. Counterweight to both “links are everything” and “links don’t matter.”

Connections

Entities mentioned: Google Concepts referenced: Local Link Building & Authority, Duplicate Content (for Service-Area Pages), Brand & User Signals as Ranking Drivers, Entity-Based SEO, Technical SEO from Architecture · E-E-A-T (NEW: e-e-a-t)

Contradictions / Tensions

  • Page-level vs. site-wide lift. Official: site-wide signals exist and do not automatically rank every page. Tension with agency talk of Domain Authority / “the site is trusted so new pages rank.” Compatible with An Anonymous Source Shared Thousands of Leaked Google Search API Documents with Me; Everyone in SEO Should See Them having host-level NavBoost/quality fields — those are site-level data, not a promise every URL inherits the boost. Flag on Brand & User Signals as Ranking Drivers.
  • PageRank still core vs. Caleb’s post-2023 “links as quality filter” (directionally compatible) vs. Dooley’s “traditional link building is diminishing / reviews replace them.” Official does not retire links; it lists PageRank among many systems and separately rewards reviews via a reviews system.
  • Exact-match domain downweight vs. keyword-in-domain folklore. Aligns with SEO Starter Guide: The Basics (URL/domain keywords barely rank).
  • Helpful content is not a separate toggle after March 2024. Do not advise “recovering from HCU” as if a named classifier still sits apart from core.
  • Passage ranking vs. Google’s later “you don’t need to chunk content for AI” (Google's Guide to Optimizing for Generative AI Features on Google Search). Compatible: Google can already score passages; you should not pre-chunk for models.

Notes

Last updated 2025-12-10 UTC. Use as the map of named systems; do not treat the list as complete or weighted. Pair with the API leak (what is stored) and this page (what Google admits is a ranking system).