On 25 October, Google announced BERT: an update it describes as the biggest improvement to search in five years. It affects around ten percent of all searches, and the impact is already measurable for many websites.

BERT stands for Bidirectional Encoder Representations from Transformers. It is an advanced natural language understanding model that helps Google grasp the context of a search query, not just each individual word in isolation.

For most SMBs, the technical details are beside the point. What matters is what it means for the content on your website.

What is actually new?

Previously, Google largely treated search terms as separate units. A search for "parking not allowed" would attempt to match pages containing those words, but whether parking was allowed or not allowed could be harder to distinguish.

With BERT, Google understands the relationship between words in a sentence: prepositions, pronouns, context. An example Google used themselves: the search "can you get medicine for someone at the pharmacy" used to be interpreted broadly as being about picking up medicine. BERT now understands that "for someone" is the critical element: the search is about collecting another person's prescription, not your own.

A seemingly small difference. But for many searches, that kind of nuance is exactly what separates a useful result from a useless one.

The practical consequence for content

BERT rewards content written to help people, and penalises content written to trick search engines.

The technique of stuffing keywords, repeating phrases unnaturally or writing sentences like "Web design London is something we at Web Design London Ltd offer for web design in the London area" is not just bad writing. It is now actively bad SEO.

Google is better than ever at understanding what a piece of text actually says, and what it is trying to say. That means:

  • Naturally and precisely written content ranks better
  • Pages that fully answer the intent behind a search rank better than pages that merely match keywords
  • Long, conversational queries (the kind we see in voice search) are better interpreted and matched

What should you do?

Nothing dramatic. But some habits are worth building:

Write for the reader, not the search engine The best advice for BERT is the best advice for content in general: write to help a person who needs an answer, not to impress an algorithm. BERT has now made these two objectives far more aligned than they were.

Answer the full question Someone searching "what should I watch out for when switching accountant?" does not just want a definition of accounting services. They want practical guidance. BERT helps Google match pages that genuinely address the full intent behind the query.

Use plain language Avoid artificial SEO prose. Write the way you would explain something to a new client on the phone. Precise, professional and human.

Do not over-optimise Keyword density, exact phrase matching and mechanical repetition are techniques from an older era. BERT rewards context and relevance, not mechanical optimisation.

A long-term trend confirmed

BERT is not a break from the past: it is an acceleration of a direction Google has been moving in for years. RankBrain in 2015 was an early step. Hummingbird in 2013 another.

The direction is consistent: Google wants to understand what people actually mean, and reward content that genuinely helps them. For businesses already writing good, expert and relevant content, BERT is good news. For those who have built visibility on technical tricks, it is a warning.

Want to assess whether the content on your website is set up for how Google interprets searches today? Get in touch: we will review your most important pages and give you concrete recommendations.