MUM, short for Multitask Unified Model, is an AI model Google announced in 2021. It can work across languages and across formats such as text and images. Google has been explicit that MUM is not currently used for general ranking in Search. Instead it supports particular applications, such as identifying variations of names in health-related searches and improving featured snippet callouts.
What makes MUM different
Google presented MUM as a step beyond BERT: trained across many languages at once, able to transfer knowledge between them, and able to take images as input. The idea was that a complex question, such as how to prepare for a hike in a different country, might draw on sources in several languages and formats.
Where it is used
Google's ranking systems guide describes MUM's use as limited to specific applications. Early examples included spotting the many names of COVID-19 vaccines across languages and improving how featured snippets highlight key information. Some visual search features have also been linked to MUM in Google's announcements.
Google has not published a full list of MUM applications, so claims that a ranking change was caused by MUM should be treated with skepticism.
Relevance for marketers
For most SaaS teams, MUM is background knowledge rather than something to act on. The broader trend it represents, search systems that read across languages and media, does support publishing clear images with meaningful context and keeping localized content accurate.
Common mistakes
Attributing ranking changes to MUM
Google says MUM is not used for general ranking. Look at core updates and other systems first.
Questions
No. Google lists BERT as an active ranking system and describes MUM as used for specific applications.
