BERT
A transformer language model Google applied to search from 2019, improving understanding of word order, prepositions and context…
GoogleSystemConfirmedHigh impact
BERT
Transformer-based language model applied to ranking and featured snippets, affecting around 10% of English queries at launch.
What changed
BERT let Google interpret the role of prepositions and word order in a query rather than treating it as a bag of keywords, which mattered most for long conversational searches. Google described it as the biggest leap forward in query understanding in five years and said it affected about one in ten US English searches, plus featured snippets in two dozen countries. It expanded to more than 70 languages in December 2019. Google said explicitly that there is no way to optimise for BERT.
Timing
Rollout began 25 October 2019. No completion date was announced. Rankings typically remain unstable for the whole rollout window, so measurements taken inside it are not a reliable read on the outcome.
What it affected
Related terminology
Google ranking systems
Google's documented list of the systems that generate rankings, distinguished from one-off 'updates' to those systems, and…
MUM
Google's multimodal, multilingual model announced in 2021, able to transfer knowledge across languages and formats to answer…
Neural matching
A Google system that uses neural networks to understand how queries relate to page concepts, bridging vocabulary gaps between…
Further reading
- Attention Is All You Need · Landmark Post
- BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding · Landmark Post
- Passage Re-ranking with BERT · Landmark Post
- Understanding Searches Better Than Ever Before · Landmark Post