What Google actually names in its own ranking systems guide

SEO discourse mostly runs on what got leaked or what a witness said under oath in an antitrust trial. Less discussed is that Google also just publishes a list. The “Guide to Google Search ranking systems,” maintained on Google’s own developer documentation, names specific systems by name and gives each one a short, plain description of its job. It is not a spec and it is not a weight table, but it is real, it is current, and almost nobody quotes it directly.

Systems the page labels as AI systems#

Five of the named systems get the same qualifier from Google: “AI system.” BERT (Bidirectional Encoder Representations from Transformers) “allows us to understand how combinations of words express different meanings and intent.” MUM (Multitask Unified Model) is described as capable of understanding and generating language, but the guide is explicit that it is “not currently used for general ranking,” only for narrower applications such as COVID-19 vaccine information searches and featured snippet callouts. Neural matching understands “representations of concepts in queries and pages” and matches them to each other. The passage ranking system identifies individual sections of a page to judge relevance to a search more precisely than page-level matching allows. RankBrain helps Google “understand how words are related to concepts,” so a page can rank for a query even without the exact words used in the search.

The distinction between RankBrain and neural matching is one Google draws itself, not one this piece is inferring. Both concern understanding meaning beyond literal keyword matching, and Google lists them as separate systems with separate one-line descriptions rather than treating one as a successor to the other.

Systems for specific ranking needs#

The rest of the guide’s current entries handle narrower jobs rather than general language understanding. Crisis information systems cover both personal crises, such as suicide or poison-related queries surfacing hotline content, and wide-scale events through the SOS Alerts system. Deduplication systems keep near-identical pages from cluttering one results page. The exact match domain system stops a domain name alone, such as one built from a keyword phrase, from earning outsized ranking credit. Freshness systems raise newer content for queries where Google judges recency matters, its own example being a just-released movie or a recent earthquake. Link analysis systems and PageRank remain listed as part of Google’s core ranking systems, in use “since Google first launched,” though the guide says how PageRank works “has evolved a lot since then.” Local news systems surface local sources through features like Top Stories and Local News. Original content systems aim to show original reporting ahead of sites that merely cite it, supported by canonical markup. Removal-based demotion systems downrank sites with a significant volume of valid legal removals or personal-information removal complaints. Reliable information systems work to surface authoritative pages, elevate quality journalism, and, when reliable information is thin, show content advisories on rapidly changing topics. The reviews system rewards “high quality reviews… written by experts or enthusiasts who know the topic well.” The site diversity system generally limits results to two listings per site, treating most subdomains as part of the root domain. Spam detection systems, the guide says, include SpamBrain by name among the tools used against spam policy violations, a detail worth flagging because SpamBrain gets no heading of its own; it is named inside the spam detection section rather than listed as a standalone entry the way BERT or RankBrain are.

What got retired, and why it matters#

The guide keeps a “Retired systems” section for exactly this reason: names change status, and Google says so rather than quietly dropping them. As of this page’s current version, four systems are listed there: Hummingbird, a major 2013 ranking overhaul; Panda, folded into core ranking systems in 2015; Penguin, aimed at link spam and folded in during 2016; and, notably, the helpful content system, which Google’s guide now describes as incorporated into its broader core ranking systems rather than continuing to operate as a separate one. Anyone still describing “the helpful content system” as a distinct, currently-running system is describing an earlier state of Google’s own documentation, not its current one.

Why we care#

This site already published a piece on NavBoost, built entirely from DOJ trial testimony and a leaked internal document, precisely because Google has never put NavBoost on a page like this one. That gap is the honest finding here: Google maintains a real, named, publicly documented list of ranking systems, and NavBoost, described under oath by its own VP of Search as “one of the important signals,” is not on it. That does not mean the guide is dishonest or that NavBoost is fabricated; it means the industry’s two best public windows into Google’s ranking systems, official documentation and trial disclosure, currently describe two non-overlapping sets of names, and treating either set as the complete picture overstates what either source actually offers. Read the guide for what Google is willing to name and describe in its own words, and read the NavBoost piece for what it takes a subpoena to surface. Neither is the full architecture, and Google’s own page says so directly: this is “some of our more notable ranking systems,” not all of them.