Google BERT Update (25 October 2019): What Changed and How to Respond

The Google BERT update applied the BERT language model to Search to help Google understand the context of words in a query. Google announced it on 25 October 2019, said it would affect about one in ten US English searches, and expanded it to more than 70 languages on 9 December 2019. It is a language-understanding change, not a quality or spam update, so there is nothing to “fix” for it, only clear writing that answers real questions.

BERT is one of the landmark changes in my full list of Google algorithm updates.

Detail	Information
Update name	Google BERT Update (Bidirectional Encoder Representations from Transformers)
Type	Language understanding (infrastructure) change, not a quality or spam update
Rollout started	25 October 2019 (announcement date; rollout was reported to have started earlier that week)
Rollout finished	Not announced; expanded to 70+ languages on 9 December 2019
Rollout length	Not published
Confirmed by Google	Yes, announced by Search VP Pandu Nayak on Google’s official blog
Official source	Google: Understanding searches better than ever before

What the BERT update changed

BERT stands for Bidirectional Encoder Representations from Transformers. It is a neural network model for natural language processing. Before BERT, much of Google’s query processing looked at words largely in sequence or as separate keywords. BERT reads a word in light of the words both before and after it, which is what “bidirectional” means. That lets it understand how small words such as “to”, “for” and “no” change the meaning of a whole query.

Google called it “one of the biggest leaps forward in the history of Search” and said: “BERT will help Search better understand one in 10 searches in the U.S. in English”.

Google’s own example

Google’s best-known example was the query “2019 brazil traveler to usa need a visa”. The word “to” is the key. The searcher is a Brazilian travelling to the USA, not an American travelling to Brazil. Before BERT, Google could miss that relationship and return results about US citizens going to Brazil. With BERT, it understood the direction of travel and returned the right information.

Another example often used to explain it is “math practice books for adults”. The word “adults” tells you who the books are for, and BERT helps Google favour results meant for adult learners rather than generic maths books for children.

Where BERT was applied at launch

• Ranking: for US English queries.
• Featured snippets: in about two dozen countries, across languages where snippets were available.
• More languages: on 9 December 2019, BERT expanded to Google Search in more than 70 languages.

How BERT differs from RankBrain

People often confuse the two. RankBrain, introduced earlier, uses machine learning mainly to help Google interpret queries it has not seen before and relate them to known concepts. BERT focuses on the context of each word within the query. They work alongside each other rather than one replacing the other, and neither can be optimised for directly.

Timeline

Date	What happened
Week of 21 October 2019	Search Engine Land reported the rollout had started earlier in the week
25 October 2019	Pandu Nayak announced BERT in Search: ranking for US English and featured snippets in about two dozen countries
9 December 2019	BERT expanded to Google Search in more than 70 languages

Because the 25 October date is the announcement rather than a confirmed first day of rollout, treat it as an approximate start.

Who was affected

Google said about one in ten US English searches were affected at launch. Despite that large share, rank trackers recorded little measurable volatility (industry observation). The reason is that BERT mainly changed results for long-tail, conversational queries: the longer, more specific searches people type or speak as full questions. Most rank tracking focuses on short head terms, where BERT made less difference.

So the sites that noticed BERT were mostly those with lots of long-tail traffic: question-and-answer content, detailed guides, forums and support pages. Some gained, because their pages answered specific questions precisely. Some lost traffic on queries where they had ranked through a loose keyword match but did not actually answer what the searcher meant.

How to tell if the BERT update affected your site

A BERT effect looks different from a core update. It is concentrated on longer queries and shows up as changes in which queries a page ranks for, rather than a general drop.

1. Segment by query length. In archived Search Console data, split queries into short (one to three words) and long (five or more words). Compare the two weeks before about 21 October 2019 with the two weeks after 25 October. A BERT effect should show mainly in the long group.
2. Look for question-style queries. Filter for queries containing words such as how, what, can, for, to, without and near. These are the queries where context words matter most.
3. Check for meaning mismatches. For long queries you lost, read the query carefully and then your page. If the page ranked because it contained the words but answered a different question, BERT is a likely explanation.
4. Check featured snippets. BERT was applied to featured snippets at launch, so gains or losses of snippets around the same dates are worth noting.
5. For non-English sites, use 9 December 2019. If your site targets another language, the relevant date for ranking effects is the expansion in December, not October.

How to recover from the BERT update

Google and the wider industry agree that there is nothing to optimise for BERT specifically, just as with RankBrain. BERT makes Google better at understanding language, so what it rewards is content that clearly answers what people mean. If you lost traffic, the answer is to match intent more precisely, not to change keyword density.

1. Write naturally for people. Use plain, conversational language. Do not force awkward exact-match phrases into sentences. BERT reads context, so natural sentences are easier for it to understand.
2. Answer the specific question early. For each important query, make sure the page gives a direct answer near the top, then the detail. This also helps with featured snippets.
3. Cover the questions around the topic. Use the questions people actually ask, from People Also Ask, your customer emails and your Search Console query report, as sub-headings with direct answers.
4. Pay attention to small words. “For beginners”, “without a prescription”, “from Pakistan to the UK”: the modifiers in a query define who it is for and what they need. If you serve that audience, say so clearly.
5. Cover the related entities. A page about choosing running shoes should naturally discuss things like cushioning, fit and durability. Covering a topic thoroughly helps Google understand what the page is about.
6. Use clear structure. Descriptive headings, short focused paragraphs and lists make it easier for both readers and Google to find the part of the page that answers a query.
7. Do not create a page for every phrasing. BERT understands that differently worded queries can mean the same thing, so one strong page usually beats many near-duplicates.

Common problem	BERT-friendly fix
Page stuffed with exact-match phrases	Rewrite in natural language with the phrase used once where it fits
Answer buried after a long introduction	Lead with a direct answer, then explain
Page ranks for long queries it does not answer	Add a section that answers them, or accept the loss if they are off-topic
Several thin pages for different wordings	Merge into one complete page

Much of this is standard on-page optimisation. If you publish in several languages, the December 2019 expansion makes this relevant to every version, which is where international SEO comes in.

Is it still relevant today?

Yes. BERT was a turning point in how Google understands language, and the direction it set has continued: Search has become steadily better at reading natural, conversational queries, and people search in longer, more specific ways, especially with voice and AI-assisted search. The practical advice from 2019 has not changed: write clearly, answer the real question, and do not try to game the wording.

If anything, the case for intent-focused writing is stronger now than in 2019, because more of Search depends on understanding meaning rather than matching keywords.

Frequently asked questions

When was the Google BERT update?

Google announced it on 25 October 2019, with rollout reported to have begun earlier that week. It expanded to more than 70 languages on 9 December 2019.

How many searches did BERT affect?

Google said BERT would help Search better understand about one in ten searches in the US in English at launch.

How do I optimise content for BERT?

You cannot optimise for BERT directly. Write naturally for people, answer specific questions clearly and early, and cover the topic thoroughly. That is what BERT helps Google recognise.

What is the difference between BERT and RankBrain?

RankBrain helps Google interpret unfamiliar queries by relating them to known concepts. BERT focuses on the context of each word within a query, especially small words that change meaning. Both work together.

Was BERT a penalty or a core update?

Neither. It was a language-understanding improvement. It did not judge content quality or target spam.

Sources

• Google Blog: Understanding searches better than ever before
• Keywords Everywhere: BERT Update
• Search Engine Journal: Google algorithm history

If you lost long-tail traffic around late 2019, or you want your pages to match how people actually search, I can review your content against the queries you should be winning. See my SEO audit services or request a free SEO audit.