Deutsch
Ranking systems

BERT

Also known as: Bidirectional Encoder Representations from Transformers

BERT (Bidirectional Encoder Representations from Transformers) is an AI system Google uses to understand how combinations of words express meaning and intent in a search query. In SEO, the term means this use in Google Search. BERT also exists as a general language model in machine learning research.

At a glance

Started
October 25, 2019
Completed
ongoing, listed by Google as an active system
Type
AI language model for ranking and featured snippets
Affects
Understanding queries, at launch about one in ten English searches in the U.S.
Confirmed by
yes

Last checked:

Google announced BERT for Search on October 25, 2019, and called it the biggest leap forward in the past five years. At launch it applied to about one in ten English-language searches in the U.S. The model reads each word in relation to the words before and after it, which helps with longer, conversational queries and with small words such as “to” or “no”. Google also used it to improve featured snippets in two dozen countries.

Google’s ranking systems guide still lists BERT as an active system.

Google ranking systems 2010–2024
  1. Google announces the completion of Caffeine, its new web indexing system. Google says it delivers 50 percent fresher results than the previous index.Source: googleblog.blogspot.com
  2. Google announces a ranking change against low-quality and copied content that noticeably affects 11.8% of queries. Google later refers to the system by its nickname “Panda”.Source: googleblog.blogspot.com
  3. Google improves how it decides when to favour recent results. According to Google the change affects roughly 35 percent of searches and builds on Caffeine.Source: googleblog.blogspot.com
  4. Google launches a change that looks at page layout. Pages with little content and many ads above the fold may rank lower.Source: developers.google.com
  5. Google announces a change against webspam, later nicknamed “Penguin”. It affects about 3.1% of English queries and roughly 3% of German queries.Source: developers.google.com
  6. Hummingbird, which Google describes as a major improvement to its overall ranking systems.Source: developers.google.com
  7. Panda becomes part of the core ranking systems. Google now lists it among its retired systems.Source: developers.google.com
  8. Google launches RankBrain, the first deep learning system used in Search. It relates the words in a query to real-world concepts.Source: blog.google
  9. Google starts rolling out the mobile-friendly update, boosting mobile-friendly pages in mobile search results. The industry unofficially calls it “Mobilegeddon”.Source: developers.google.com
  10. Penguin becomes part of the core algorithm and now works in real time. Before that, the list of affected sites was only refreshed periodically.Source: developers.google.com
  11. Google introduces neural matching. The system helps relate queries to pages through underlying concepts.Source: blog.google
  12. Launch of SpamBrain, Google’s AI-based spam prevention system. Google says it identified nearly six times more spam sites in 2021 than in 2020.Source: developers.google.com
  13. Google applies BERT to ranking and featured snippets. At launch it affects about one in ten English searches in the US.Source: blog.google
  14. Google presents passage ranking, which can assess individual passages of a page for a query. Google expects it to improve 7 percent of queries.Source: blog.google
  15. First product reviews update. It rewards in-depth product reviews and initially covers English-language reviews only.Source: developers.google.com
  16. Google introduces MUM, a model trained across 75 languages. Google calls it 1,000 times more powerful than BERT.Source: blog.google
  17. First link spam update. It aims to identify and nullify link spam across multiple languages and rolls out over two weeks.Source: developers.google.com
  18. With the helpful content update, Google aims to show more original, helpful content written for people.Source: status.search.google.com
  19. December 2022 helpful content update. Google says it improves the classifier and works globally in all languages.Source: status.search.google.com
  20. The product reviews update becomes the reviews update. Google says it now covers reviews of products, services and other things.Source: status.search.google.com
  21. September 2023 helpful content update. Google says it improves the classifier again.Source: status.search.google.com
  22. November 2023 reviews update. As the reviews system is now improved at a regular pace, Google no longer posts future updates on the Status Dashboard.Source: status.search.google.com
  23. The helpful content system becomes part of the core ranking systems. Google now lists it among its retired systems.Source: developers.google.com
  24. Three new spam policies against expired domain abuse, scaled content abuse and site reputation abuse. The March 2024 spam update launches the same day.Source: developers.google.com
  25. The site reputation abuse policy takes effect. Google had given site owners time to prepare since March.Source: developers.google.com
  26. In the US antitrust case against Google, the court describes Navboost as a signal that memorises which documents users click for a query. This is a court document, not a Google publication.Source: storage.courtlistener.com
  27. Google tightens the site reputation abuse policy. First-party involvement does not change the assessment when third-party content exploits the host site’s ranking signals.Source: developers.google.com

A selection, not a complete list. Core and spam updates since November 2021 have their own timeline. Names such as Florida (2003), Pigeon (2014), Possum (2016), Fred (2017) and Medic (2018), as well as Payday Loan and EMD, come from the SEO industry. Their dates rest only on trade press reports and posts by Google employees on X; Google has not published a dated page about them. Last checked: .

Example

For the query “2019 brazil traveler to usa need a visa”, earlier systems returned results about U.S. citizens traveling to Brazil. With BERT, Google recognized that “to” marks the direction of the trip, so the results cover a Brazilian traveling to the U.S.

FAQ

Does Google still use BERT? Yes. The ranking systems guide lists it among the active systems.

When was BERT introduced in Search? October 2019.

Sources

  1. Understanding searches better than ever before – Google Blog, 2019-10-25blog.google
  2. Ranking systems guide – Google Search Centraldevelopers.google.com

Last updated: