What is BERT?
BERT stands for Bidirectional Encoder Representations from Transformers, which is a mouthful for a simple idea: read a sentence in both directions at once, so every word is understood in the light of the words around it. Before BERT, search handled a query largely as a bag of keywords. After it, Google could tell that "can you get medicine for someone pharmacy" is a question about picking up a prescription for another person, because the word "for" changes the meaning. Google called it the biggest leap forward in search in five years and said it affected about one in ten English queries at launch.
It was not a penalty update and it did not target sites. It was Google getting better at reading.
Why it matters
Why BERT matters
Because it marked the point where search stopped rewarding keyword matching and started rewarding meaning. Prepositions, word order and small words that earlier systems ignored became part of what Google understood, which meant pages that answered the actual question outranked pages that merely contained the words. For anyone writing for the web, the lesson was blunt: there is no optimizing for BERT except by being clear about intent.
It also raised the ceiling on what a search could be. Longer, conversational, specific queries that used to return junk began returning good answers, and people adjusted their habits accordingly. The models that followed, from MUM to the systems behind AI Overviews, build on the same foundation.
In practice
How BERT works in practice
- 01
BERT runs on the query and on passages of pages, scoring how well a passage matches what the person meant. It is one signal among many in a ranking system that also weighs links, freshness, page experience and dozens of other factors, so it shapes which pages are judged relevant rather than deciding everything on its own. Featured snippets were one of the first places its effect showed.
- 02
For a content team the practical response is to write the way people ask. Use natural sentences, answer the question in full, keep the context on the page rather than assuming the reader already has it, and stop shaving prepositions out of headlines to make a keyword fit. Writing well and writing for BERT turned out to be the same task.
Watch out
What people get wrong about BERT
Trying to optimize for it. There is no BERT checklist, no tag to add, no density to hit, and Google's own advice at launch was that there was nothing to do except write for people. Agencies that sold "BERT optimization" were selling the weather. The other mistake is assuming it made keywords irrelevant: people still search in words, and the page still needs to use them. BERT just means using them in sentences that mean something.
Treating it as current. BERT was a milestone, not the state of the art. Search has moved through several larger models since, and the principles it introduced are now simply how search works.
