MUM (Multitask Unified Model)

MUM (Multitask Unified Model) is an AI model Google announced in May 2021 that can understand and generate language, work across many languages and handle information in more than one format, such as text and images. Google uses it for specific search tasks rather than ranking every query.

Written by Paul Hoda, SEO consultant since 2005. Last updated 8 November 2026.

What MUM is

Google introduced MUM at its I/O conference in 2021. Like BERT, it is built on the transformer architecture, but Google described it as far more powerful. Google highlighted three main abilities:

  • Multitask: it can perform several tasks at once, such as understanding a question and summarising information.
  • Multilingual: Google said it was trained across 75 languages, so it can transfer knowledge learned in one language to another.
  • Multimodal: it can understand information from text and images, with Google mentioning future potential for video and audio.

The problem MUM was designed to solve

Google’s launch example was a hiker who has climbed Mount Adams and wants to know how to prepare to hike Mount Fuji next autumn. Answering that would normally take many separate searches about elevation, weather, gear and training. Google said MUM could understand the comparison being made and bring together relevant information, including from sources in other languages such as Japanese.

How Google has used MUM

Google has been clear that MUM is applied to particular tasks rather than being a general ranking system. Uses Google has described include:

  1. Vaccine searches: Google said MUM helped it identify hundreds of names for COVID-19 vaccines across many languages, so people could find reliable information quickly.
  2. Crisis information: improving detection of searches from people in personal crisis so helpful resources could be shown.
  3. Featured snippets: Google said MUM helped improve the quality of snippet callouts.
  4. Visual search: Google described plans to combine images and text in multimodal search, such as searching with a photo and adding a question.
  5. Related topics: features designed to help people explore a subject more broadly.

What MUM means for SEO

There is no MUM update to recover from and nothing to optimise for directly. MUM is part of a long trend in which Google’s systems understand topics, languages and formats more like people do. Practical implications include:

  • Covering topics in depth, including the follow-up questions people naturally ask.
  • Using descriptive images with good alt text, since visual content is increasingly understood.
  • Providing original information that adds something new, as AI systems can draw on content from many languages.
  • Organising content clearly so specific passages can be understood on their own.

MUM in context

MUM sits between BERT and the large language models that now power Google’s AI features, including the Gemini models behind AI Overviews. It showed Google moving from understanding queries towards bringing together information to answer complex needs, a direction that has since become central to AI search.

Key Takeaways

  • MUM is a Google AI model announced in May 2021.
  • It is multitask, multilingual and multimodal.
  • Google uses it for specific tasks, not for ranking every query.
  • Examples include vaccine information and improving featured snippets.
  • Thorough, original, well-organised content suits systems like MUM.

Related Terms

Want help putting this into practice? Get a free SEO check, or keep learning in the SEO Encyclopedia.