Fundamentals
Chapter 03 / 09
The Google algorithm
It's not one algorithm — it's a stack of overlapping ranking systems updated continuously, with five core update windows a year that move pages 30 positions overnight. Here's the actual model.
“The Google algorithm” is shorthand for at least five overlapping ranking systems running continuously,with named “core updates” rolling out four to six times a year that visibly move rankings 10-30 positions overnight. Treating it as a single black box is the reason most teams misdiagnose ranking drops and chase fixes that don’t address the actual cause.
This article unpacks the actual stack — what each system does, how core updates differ from continuous tuning, and which signals are doing the heavy lifting in 2026.
“Pages don’t fail because of “the algorithm.” They fail at one specific layer of it. Naming the layer is the diagnosis.”
The five overlapping systems
Google has confirmed five distinct ranking systems publicly through its Search Liaison + Webmaster Central communications. Each one handles a different scoring question. They run together on every query, and the final rank is a composite score across all of them.
| System | Question it answers | How it updates |
|---|---|---|
| Core ranking | Does this page deserve to rank, given the query? | Continuous + ~5 named Core Updates per year |
| Helpful Content | Is this content useful to a real human, or written for an algorithm? | ~2-4 Helpful Content Updates per year |
| Spam (SpamBrain) | Is this site trying to manipulate rankings via banned tactics? | ~3-5 Spam Updates per year + continuous |
| Language models (BERT / MUM / Gemini-grounded) | What does the query actually mean? What does the page actually say? | Continuous, no announcements |
| AI Overview synthesis | Which sources should be cited inside the AI-generated answer? | Continuous, separate from organic ranking |
1. Core ranking — the base score
The original PageRank-derived ranking system, evolved across 25 years. Scores pages on hundreds of factors — relevance, authority, freshness, intent match, technical health — and produces the base score that the other systems modify.
“Core Updates” are the named, announced reweightings of this base system. Google confirmed five of them in 2025 (March, June, August, November, December). Each one shifts which factor categories carry more weight: a 2024 update favoured original reporting; a 2025 update favoured editorial depth over programmatic volume. The page didn’t change; the score weights did. Pages that were over-relying on the now-devalued factor drop, sometimes by 30 positions.
2. Helpful Content System — the human filter
Launched 2022, integrated into core ranking 2024. Asks whether a page reads as written for a real human or assembled to game search. Site-wide signal — a single page that fails can drag down the rest of the site’s rankings.
What it down-weights: pages built around keyword density without addressing the actual user need; AI-generated content that adds no new perspective; thin content that summarises Wikipedia; over-templated programmatic pages. What it rewards: original analysis, first-hand experience signals, content that answers the actual question rather than the keyword variant.
3. SpamBrain — the manipulation filter
Google’s machine-learning spam classifier. Catches paid link networks, scaled content abuse, expired-domain abuse, doorway pages, hidden text, cloaking. Triggers manual actions visible in Search Console. Recovery from a manual action takes 1-6 months and a successful reconsideration request — there is no shortcut.
4. Language models — what the page says vs what the query asks
BERT (2019) and MUM (2021) were the first transformer-based systems Google deployed in ranking. They handle the semantic match between query and page — “is this page actually about what was asked?” — far more accurately than the keyword-overlap matching that preceded them.
In 2025 Google added Gemini-grounded ranking layers, which now also score whether the page’s entities are recognised against Google’s knowledge graph and whether its claims are consistent with corroborating sources. This is the layer that punishes “AI-generated content with confident wrong facts” — the model checks claims against known sources and de-weights pages that fail.
5. AI Overview synthesis — the citation engine
Separate from organic ranking. Decides which sources to pull and quote when composing an AI Overview answer. Scores pages on different factors: passage-level self-containment, entity clarity, citation-readiness of facts, freshness for time-sensitive queries.
A page can rank position 8 in organic and be the most-cited source in the AI Overview for the same query. Or rank position 3 organic and never appear in the overview. The two systems share inputs but score them differently. How search engines work covers the synthesis stage in more detail.
What signals actually move the needle in 2026
Of the often-cited “200+ ranking factors,” six categories carry disproportionate weight in 2026:
- Search intent match.Does the page format match what the SERP shows for the query? Informational queries that return guides won’t rank a product page no matter how good. The fastest way to misdiagnose: optimise on-page elements while the page format is the real mismatch.
- Content depth + originality. Pages with first-hand analysis, original data, or expert perspective outrank summaries of the same topic by competitors. The Helpful Content System weights this heavily.
- Backlink quality. One link from a recognised editorial site outweighs 1,000 links from low-quality directories. The opposite of the 2010s, when volume mattered.
- Brand entity strength. sameAs identity across LinkedIn / Wikipedia / Crunchbase / X, plus citation count in trusted media. The Gemini-grounded ranking layer reads this directly.
- Core Web Vitals + technical health. A tiebreaker, not a primary lever. Two pages otherwise tied: the faster, cleaner one wins.
- Freshness for time-sensitive queries.“Best CRM 2026” needs a 2026-dated update; “how to write a will” doesn’t.
How core updates actually unfold
A typical Google Core Update runs on a predictable arc: announcement (rare — usually a Search Liaison tweet), 1-3 week rollout window where SERPs go volatile, plateau where new rankings settle, then 4-6 weeks of community analysis until the broad pattern is named.
The mistake teams make: panicking inside the rollout window and shipping changes while the SERP is still moving. By the time you ship, the update has reached your URL and your changes are competing against a moving baseline. Wait for the plateau, diagnose the actual layer that lost weight, fix the substrate.
The triage flow
| Symptom | Likely system | Likely fix |
|---|---|---|
| Sudden drop after a named Core Update | Core ranking re-weight | Identify which factor was devalued (read Search Liaison + community analysis); strengthen the substrate, not the page |
| Site-wide demotion across many pages | Helpful Content System | Audit content quality + thin pages; the system penalises sites, not pages |
| Specific pages disappear from index | SpamBrain manual action | Check Search Console > Manual Actions; submit reconsideration once root cause is fixed |
| Page ranks but answers wrong query | Language model intent match | Restructure page to match the actual query intent (informational vs transactional) |
| Ranks in organic, never appears in AI Overview | AI Overview synthesis | Restructure passages to be self-contained; tighten entity description |
Common questions
Common questions
Quick answers to what we get asked before every trial signup.
The continuous ranking systems are updated daily — small adjustments to RankBrain, BERT, Gemini-grounded ranking, and the spam classifiers happen every day with no announcement. Confirmed Core Updates roll out four to six times a year, take 1-3 weeks to fully deploy, and are the ones that move rankings 10-30 positions visibly. Helpful Content and Spam Updates land separately, usually 2-4 times a year each.
In this cluster