Google MUM

How MUM changed Search

Google announced MUM at I/O 2021 and framed it as a way to handle questions that take several searches today. Its own example describes someone who has hiked Mt. Adams and now wants to hike Mt. Fuji next fall, wanting to know what to do differently to prepare. A person would normally run many queries (elevation comparison, gear, seasonal weather). MUM is meant to connect those threads in one pass, since it is trained on 75 languages at once and can transfer knowledge between them. Google also says MUM “not only understands language, but also generates it.”

Its first real application was practical, not flashy. Google used MUM to identify over 800 vaccine name variations across more than 50 languages “in a matter of seconds,” then surfaced trustworthy COVID-19 information. Later MUM-powered features reached the visible SERP and Lens: multimodal point-and-ask image-plus-text queries, deeper video topic understanding, and the “Things to know” and refine/broaden panels.

For SEO, MUM matters as a direction, not one ranking update. It marks Google moving toward multimodal, multilingual, intent-first understanding, the same trajectory that later produced AI Overviews. Practical takeaways: write for the whole task, not a single keyword; add real images and structured context; and cover the follow-up questions a searcher asks next. Google has never published a “MUM score,” so treat it as machine-reading capability, not a checklist.