You cannot diagnose this with certainty, and being upfront about that limitation is the actual first step of the diagnostic. Google does not expose model-level attribution anywhere in Search Console, ranking documentation, or any public tool. There is no report that says “this ranking movement is attributable to MUM.” What you can do is observe SERP composition changes for a query set over time and treat increased multimedia prominence as a directional signal about content-format preference for that query type, not as proof that a specific named system caused it.
Why direct attribution isn’t possible
Google introduced MUM (Multitask Unified Model) at I/O 2021 and described it in Search Central and Google blog posts as a model roughly 1,000 times more powerful than BERT, capable of understanding information across text, image, and (eventually) video and audio simultaneously, and across languages. Google’s own public statements about MUM have consistently focused on a small number of disclosed applications: things like improving how Search understands complex, multi-part queries, surfacing related-topic suggestions, and specific features like the “things to know” expansions and visual-search-adjacent experiences. Google has never published a comprehensive statement saying MUM is now a general-purpose ranking factor that systematically rewards multimedia-rich pages across the board.
This matters diagnostically because ranking systems at Google are layered: there are hundreds of signals, dozens of named systems (helpful content system, passage ranking, various neural matching components, and so on), and no public documentation maps a given SERP change to a specific system’s internal weight adjustment. Even Google’s own search liaison, when asked about ranking volatility, routinely says core updates and system interactions are complex and not reducible to a single cause. If Google itself doesn’t attribute individual SERP movements to individual named systems in public communication, an outside practitioner attempting to isolate “this is MUM specifically” is working with less information than Google has and still can’t do it. Any tool or consultant claiming to “detect MUM impact” is fabricating a level of certainty that doesn’t exist in the available data.
What you’re actually able to observe is correlational and format-level: are queries in a given cluster increasingly showing image packs, video carousels, or a mix of rich results next to organic listings? Is the organic slot itself trending toward results that embed video or substantial imagery? That’s a real, measurable pattern. Calling it “MUM” is a labeling choice, not a verified causal claim.
What you can actually observe and test
Since true attribution isn’t available, the useful diagnostic work shifts to SERP-observation methodology and controlled content-format testing.
SERP feature auditing over time. For a defined set of target queries, track the presence and position of image packs, video thumbnails inline in organic results, “videos” carousels, and any multimedia-rich featured snippets. Do this as a longitudinal log (weekly or monthly snapshots), not a single point-in-time check, because a single snapshot tells you nothing about trend direction. If you see a consistent increase in multimedia SERP features for a query cluster across several months, that’s a real, defensible observation about how Google is choosing to answer that query type right now, regardless of which internal system is responsible.
Segmenting by query intent. Multimedia-favoring shifts tend to cluster around specific intents (how-to, visual/product-comparison, entertainment, tutorial-style informational queries) rather than appearing uniformly across all query types. If you only see the pattern in visually-oriented or instructional queries and not in, say, definitional or transactional queries, that’s useful evidence the shift is intent-driven rather than a blanket multimedia preference, which is a more defensible read than attributing it to a single model.
Before/after content-format testing on your own pages. Where you control the page, you can test format changes directly: add a genuinely relevant embedded video or original imagery to a page that previously had text only, and track ranking and impression changes for the target query over a period long enough to rule out normal volatility (generally several weeks minimum, since short windows are dominated by noise). This doesn’t prove “MUM” did anything, but it gives you a real causal signal about whether format changes affect performance for your specific pages and queries, which is the actionable question anyway.
Cross-referencing with known update timing. Compare any SERP-composition shift against Google’s publicly announced core updates, spam updates, and other confirmed rollouts (tracked via Google’s Search Status Dashboard and Search Central update history). If a multimedia-favoring shift coincides with an announced core update, you can reasonably say “this shift correlates with a confirmed algorithm update” rather than naming an unconfirmed internal cause. That’s a meaningfully more honest claim than “MUM did this.”
Competitive format gap analysis. Look at what’s actually ranking for your target queries right now: are top-ranking pages predominantly combining text with video, images, or interactive elements versus text-only competitors? If there’s a consistent pattern across the top results, that’s useful competitive intelligence about current format expectations for that query, independent of why Google is currently favoring it.
The honest framing to operate from
Treat “MUM caused this” as an unfalsifiable hypothesis with current public tools, and don’t build a client-facing narrative around it. What’s defensible is: “SERPs for this query cluster are increasingly showing multimedia-rich results, consistent with Google’s broader, publicly disclosed investment in multimodal understanding since 2021, though we can’t isolate which specific system is responsible.” That framing lets you act on the pattern (build genuinely useful multimedia content where it serves the query) without overselling a mechanism you can’t verify. The practical output is the same either way: pages that authentically combine well-produced text, relevant original imagery, and video where it adds real value to the query tend to perform better for visually- or instructionally-oriented searches. You don’t need to prove MUM is the reason to justify doing that work; the SERP-observation and format-testing data justifies it on its own.
Hypothetically, imagine an SEO team notices that for a cluster of “how to fix” instructional queries, image packs and video carousels have become noticeably more prominent in the SERPs over the past several months, call this hypothetical client “Example Home Repair.” If the team’s report to the client said “MUM is now favoring your competitors’ video content,” that would be presenting an unverifiable causal claim as fact. A more defensible version, in this same hypothetical, would document the SERP-composition trend month over month, note that it clusters specifically around instructional queries rather than definitional ones, and recommend adding genuinely useful embedded video to Example Home Repair’s top instructional pages, tracking the actual before/after ranking and impression data over several weeks rather than attributing the shift to a specific named model.
