SEO

BERT & MUM (Google NLP models)

Q: What are BERT and MUM?

BERT (2019) and MUM (2021) are Google language models that revolutionised the understanding of search queries. BERT understands context bidirectionally, MUM is multimodal and 1,000x more powerful. They form the basis for Google's current understanding of search intent.

Q: What changed with BERT?

Google understands conversational searches. Before BERT: ”Can I pick up medication for someone at the pharmacy?” was interpreted incorrectly (focus on individual keywords). After BERT, the ”for someone” is understood in context.

Q: Is MUM still active today?

Yes, in various Google systems. MUM feeds Google Multisearch, Google Lens and featured snippet generation. It has been developed further into newer language models, with Gemini on the way as the successor for many MUM tasks.

Q: What does this mean for content creation?

Write naturally, not for algorithms. Keyword stuffing is obsolete — Google understands semantic relationships. Clear, natural language with topical coherence ranks better than optimised keyword text.

Also known as: BERT, MUM, Google BERT, Google MUM, Natural Language Processing

BERT (Bidirectional Encoder Representations from Transformers, published by Google in 2018, live in search from 2019) and MUM (Multitask Unified Model, announced in 2021) are Google's own families of language models, which have dramatically improved language understanding in the search engine. BERT understands the context of a word within a sentence better than the old keyword-matching algorithms. MUM combines this with multi-modal understanding (text + images + video) and the ability to transfer knowledge across languages. Both form the basis for Google interpreting search intent considerably better today — long tail queries, indirect phrasing, ambiguous phrases.

What BERT understanding means in practice

Example query: ”2019 brazil traveler to USA need a visa”. Before BERT (as of: up to 2019), Google did not understand the word ”to” in the right context and often showed results about ”US citizen travels to Brazil”. After BERT, the system understands: the traveller is Brazilian, the destination is the USA — and delivers relevant visa information. Effect for SEO: literal keyword stuffing has less impact, semantic topical completeness more.

MUM as an extension

MUM goes beyond BERT: it combines understanding from text, image and video, and it can transfer knowledge between languages. For example, a query in English about ”best hiking boots for Patagonia trekking” can be answered by MUM using information from Spanish sources about Patagonia — translated automatically. Effect for SEO: English content visibility benefits from high-quality non-English content in the same topic area.

What BERT/MUM mean for optimisation

Natural language instead of keyword stuffing — Google understands the context.
Search intent match matters more than ever — exact keyword matching helps nothing if the intent does not fit.
Complete answers to the search question instead of SEO tricks.
Long tail clusters benefit disproportionately — Google connects long tails with semantically related main topics.
Multilingual content becomes more valuable — good original language plus clean hreflang markup.

Example from practice

Example: In 2019, a marketing website still had many keyword-stuffed texts (”online marketing cheap online marketing Berlin cheap marketing online Berlin”). After the BERT roll-out, visibility declined systematically because naturally written competitor texts were understood better. A switch over 8 months to natural language, clear H2 structures and semantic topical completeness meant visibility not only recovered but grew by 60 % above the pre-BERT level. A classic case of successful adaptation to language models.

Frequently asked questions

What are BERT and MUM?

BERT (2019) and MUM (2021) are Google language models that revolutionised the understanding of search queries. BERT understands context bidirectionally, MUM is multimodal and 1,000x more powerful. They form the basis for Google's current understanding of search intent.

What changed with BERT?

Google understands conversational searches. Before BERT: ”Can I pick up medication for someone at the pharmacy?” was interpreted incorrectly (focus on individual keywords). After BERT, the ”for someone” is understood in context.

Is MUM still active today?

Yes, in various Google systems. MUM feeds Google Multisearch, Google Lens and featured snippet generation. It has been developed further into newer language models, with Gemini on the way as the successor for many MUM tasks.

What does this mean for content creation?

Write naturally, not for algorithms. Keyword stuffing is obsolete — Google understands semantic relationships. Clear, natural language with topical coherence ranks better than optimised keyword text.

Used in Rankmio for

Semantic audit in the SEO check

Go to the feature →

Last updated: 2026-06-17 · Browse all glossary entries