--- type: "Article" title: "BERT Google | topical authority content" description: "BERT in Google Search: what Google confirms, how it reads word order and small words, and a practical check for query direction mismatches." resource: "https://www.laurelinlabs.com/search-systems/bert" tags: ["bert-google", "topical-authority-content", "bert-natural-language-understanding-search"] generated: { by: "laurelin-site-build/2026-09-02", at: "2026-09-02T12:00:00+00:00" } status: "stable" stale_after: "2027-03-02T00:00:00+00:00" --- # BERT BERT is a transformer based language model, published by Google researchers in 2018, that Google Search uses to understand how combinations of words express meaning and intent. Google confirms it in its ranking systems guide and said in 2020 that BERT was used in almost every English query, for both ranking and retrieval. It rewards nothing a site can add as markup. What it changes is that words like to, for and no now count, so pages must answer the query as actually phrased. ## What is BERT? BERT (BERT) is a language model that reads each word in the context of the words on both sides of it. Google researchers [published it in October 2018](https://arxiv.org/abs/1810.04805), and Google began applying it to Search in [October 2019](https://blog.google/products-and-platforms/products/search/search-language-understanding-bert/). In Search, Google uses it to work out what a query means when word order and short function words matter. ## How does BERT change how Google reads a query? Earlier systems leaned on matching individual keywords. BERT considers the full sentence, so prepositions and negations are read rather than discarded. Google's [2019 launch post](https://blog.google/products-and-platforms/products/search/search-language-understanding-bert/) used the query "2019 brazil traveler to usa need a visa", where the word "to" decides which direction of travel the searcher means. Google later said BERT is used in both ranking and retrieval. In [February 2022](https://blog.google/products-and-platforms/products/search/how-ai-powers-great-search-results/) Pandu Nayak wrote that Google had also improved legacy systems with BERT training, and that BERT works as part of an ensemble rather than alone. ## What evidence is there that Google uses BERT? The evidence label is **Confirmed**. Google names BERT in its [ranking systems guide](https://developers.google.com/search/docs/appearance/ranking-systems-guide) and has announced each step of its roll out. At launch in [October 2019 Google said BERT would help with one in 10 English searches in the US](https://blog.google/products-and-platforms/products/search/search-language-understanding-bert/). A year later, at [Search On in October 2020](https://blog.google/products-and-platforms/products/search/search-on/), Google said it was used in almost every English query. In [March 2022 Google said a BERT improvement had cut unexpected shocking results by 30%](https://blog.google/products-and-platforms/products/search/using-ai-keep-google-search-safe/) over the previous year. ## What should sites do about BERT? There is no BERT markup or setting. The practical consequence is narrower than many guides suggest: small words in a query now change which page is the right answer. Look at the queries a page earns that contain direction or negation words such as to, from, for, without and no. If the page answers the opposite case, for example travel from a country when the query asks about travel to it, BERT makes that mismatch more visible, not less. Fix the answer, or build the page the query actually needs. ## Which Laurelin audit checks test for BERT? These checks measure whether pages answer the query as it is phrased. They are proxies; no audit can see BERT's output. [Page answers the reverse of the query's direction or negation](https://www.laurelinlabs.com/audit/topical-content/332-page-answers-the-reverse-of-the-query-s-direction-or-negation): the new check for this page, testing queries where a small word flips the meaning. [Content not matching search intent](https://www.laurelinlabs.com/audit/topical-content/168-content-not-matching-search-intent): the general form of the same failure. [Question heading with no direct answer](https://www.laurelinlabs.com/audit/topical-content/157-question-heading-with-no-direct-answer): BERT improved featured snippets, which need a direct answer to extract. [Answer buried, lead doesn't answer](https://www.laurelinlabs.com/audit/topical-content/172-answer-buried-lead-doesn-t-answer): the answer to the exact phrasing should appear early. [Poor auto-translated content](https://www.laurelinlabs.com/audit/topical-content/176-poor-auto-translated-content): Google applies BERT across languages, and machine translation often breaks the small words BERT reads. ## What are the key dates for BERT Google? - 2018-10-11: BERT paper first published on arXiv ([source](https://arxiv.org/abs/1810.04805)) - 2019-10-25: Google announces BERT in Search for one in 10 US English queries ([source](https://blog.google/products-and-platforms/products/search/search-language-understanding-bert/)) - 2020-10-15: Google says BERT is used in almost every English query ([source](https://blog.google/products-and-platforms/products/search/search-on/)) - 2022-03-30: Google reports BERT cut unexpected shocking results by 30% ([source](https://blog.google/products-and-platforms/products/search/using-ai-keep-google-search-safe/)) ## FAQ ### Does Google still use BERT? Yes. BERT is listed as a current system in Google's ranking systems guide, and Google said in 2020 that it was used in almost every English query. ### Can you optimise content for BERT? There is nothing BERT specific to add. The useful work is checking that pages answer the query as phrased, including direction and negation words. ### Is BERT the same as MUM? No. MUM is a later, larger model that can also generate language, and Google says it is not used for general ranking, whereas BERT is. Hub: [Topical Content](/topics/topical-content.md)