Google algorithm names

The algorithm Google was built on: a score for each page based on the quantity and quality of pages linking to it, named after co-founder Larry Page. The public toolbar PageRank score was retired in 2016, but the internal version remains a foundational ranking signal. The Reasonable Surfer patent (filed 2004, refiled 2012) extended the original model: links pass different amounts of PageRank depending on prominence, position, anchor text, and likelihood of being clicked.

Google's first ranking signal for site speed: initially desktop-only and described as affecting less than 1% of queries. In practice the bar was low: only genuinely slow sites saw a demotion. The signal evolved through PageSpeed Insights and Lighthouse before being subsumed by the Core Web Vitals in the Page Experience update.

An infrastructure update rather than a ranking change. Google rebuilt its indexing system on top of a new architecture (Percolator) that could update the index continuously instead of in batches. Fresh content reached the index ~50% faster after Caffeine. Without it, the Freshness and Panda updates that followed wouldn't have been mechanically possible.

The first major content-quality algorithm, aimed at content farms, thin pages, and duplicate or scraped content. Hit 11.8% of English queries at launch. Google trained Panda by having human raters score sites for quality, then used machine learning to find the signals those scores correlated with. The shortlist Google published shortly after became the template for every content-quality framework since. Panda ran as a periodic refresh until January 2016, when it was folded into core ranking.

Built on the older Query Deserves Freshness (QDF) concept from 2007: the idea that some queries (breaking news, recurring events, evolving topics like 'best phones') want recent results, and others don't. The November 2011 update affected ~35% of searches, more than Panda. It's the system behind the date-stamped results you see for trending queries.

Also known as the 'top-heavy' update. Penalised pages that pushed content below the fold by stacking ads at the top of the page. Narrower in scope than Panda: it ran as a separate filter that refreshed periodically rather than a continuous signal. The final refresh ran in February 2014; after that it was folded into core ranking.

Google's response to the link-building industry that had grown up around PageRank: sites buying, exchanging, and spinning links to inflate rankings. The original Penguin was a periodic filter: sites hit by it had to wait for the next refresh to recover. Penguin 4.0 (September 2016) folded the algorithm into core ranking and made it real-time, devaluing spammy links on the spot rather than penalising the site. From 4.0 onward, the Disavow tool became the recovery path of last resort rather than first.

A spam-targeted update aimed at queries Google described as 'very spammy': payday loans, casinos, pharmaceuticals, adult content. Version 2.0 followed in May 2014 (targeting the sites themselves), and 3.0 in June 2014 (targeting the queries). Affected ~0.3% of US queries at launch, but up to 4% in markets like Turkey where the spam was more pervasive.

Not a tweak, but a wholesale rewrite of Google's core ranking algorithm. Think of Hummingbird as the engine, with everything else (Panda, Penguin, RankBrain, BERT) bolted on top. The headline shift was support for conversational search. Google could finally parse the meaning of a full natural-language question rather than treating queries as a bag of keywords. Google ran Hummingbird live for a month before announcing it on the company's 15th birthday.

A rewrite of local search to make Maps and Local Pack rankings reflect the same signals as standard web search. The radius of 'local' tightened significantly. A query without a city name would surface results much closer to the searcher than before. US-only at launch; rolled out to the UK, Canada, and Australia in December 2014.

Mobilegeddon. Google's first ranking signal that demoted pages on mobile searches if the page itself wasn't mobile-friendly. Applied page-by-page, not site-wide, and only on mobile SERPs. The impact in practice was smaller than the name suggested. Google had given two months' notice and most large sites had already responded.

Google's first major machine-learning ranking signal, focused on long-tail and never-seen-before queries (around 15% of daily traffic at launch). In 2016 Google described RankBrain as one of the top three ranking signals alongside links and content. It's been part of the core stack ever since, though Google rarely names it now.

Google's AI-based spam detection system. Launched internally in 2018, named publicly in April 2022 as part of the annual webspam report. Powers Google's named spam updates, including the June and August 2021 Link Spam Updates and the December 2022 Link Spam Update, which took 29 days to roll out. In 2021, SpamBrain identified 6× more spam sites than the year before, cutting hacked spam in results by 70%.

A change to how Google crawls and indexes the web rather than a ranking signal. Googlebot's mobile crawler became the primary source of indexing, with desktop content as a fallback. Announced in November 2016, the rollout started in March 2018 for individual sites and was effectively complete by July 2024 when Google stopped indexing sites that didn't load on mobile. For responsive sites this changed nothing. For sites with separate `m.` subdomains or stripped-down mobile versions, it meant Google was now indexing the cut-down version.

Industry nickname for the August 2018 Core Update, given because medical and other YMYL (Your Money, Your Life) sites saw the biggest swings. It wasn't specifically a 'medical' update (the same algorithm ran across all queries), but the YMYL skew was sharp enough that the nickname stuck. This was the update that made E-A-T (Expertise, Authoritativeness, Trustworthiness, later E-E-A-T) a permanent fixture of SEO conversations.

The SEO industry's nickname for the March 2019 Core Update, a reference to the original Florida update of November 2003. Like all broad core updates, Google offered no specifics on what changed. Sites that lost visibility were directed at the quality rater guidelines.

Bidirectional Encoder Representations from Transformers. At launch, Google called it the largest leap in understanding queries in five years. BERT processes a query in the context of every other word in the sentence rather than reading left-to-right, so it can finally hold onto small prepositions (`to`, `for`, `from`) that flip the meaning of a long-tail search. The initial rollout affected ~10% of English queries; by December 2019 BERT was running across 70 languages.

An attempt to reward reviews built on first-hand testing and original research, and demote thin affiliate pages that recycled manufacturer specs. The first rollout was English-only and finished 22 April. Google released eight further product reviews updates over the next two years before folding the system into the core ranking systems in late 2023.

Multitask Unified Model. Google's successor architecture to BERT, announced at Google I/O 2021 and described as 1,000× more powerful. MUM is multilingual (75 languages) and multimodal, meaning it can reason across text and images at the same time. It powers specific search features (the COVID-vaccine-name normalisation in 2021, AI Overviews from 2023 onward) rather than running as a single ranking update.

The first ranking system to use the Core Web Vitals (`LCP`, `FID`, and `CLS`) as direct inputs, combined with mobile-friendliness, `HTTPS`, and the existing intrusive-interstitial signal. Mobile rollout started 15 June 2021 and completed in late August. Desktop followed in February 2022. In practice the ranking impact was modest; Google described it as a tiebreaker among pages of similar relevance. `FID` was retired and replaced by `INP` in March 2024.

A site-wide classifier targeting content written for search engines rather than people. Unlike Panda, which acted at a page level, the Helpful Content signal could pull down rankings for an entire site if enough unhelpful content was present. The fix required cleaning up or removing the offending content sitewide. The first rollout took 15 days. Google folded the system into core ranking signals with the March 2024 Core Update.

Google called this the most consequential core update they'd ever shipped, with a goal of cutting low-quality, unoriginal content in results by 40%. It rolled out alongside three new spam policies: expired domain abuse, scaled content abuse, and site reputation abuse (the 'parasite SEO' policy that took effect 5 May 2024). The core update itself took 45 days to roll out, finishing 19 April. The Helpful Content system was folded into core ranking with this release rather than running as a separate signal.