Embedding Proximity Flaws Bleed Irrelevant Click Data into Rankings via 'Now Boost'

I observe that minor, unannounced algorithm updates trigger "embedding to embedding" proximity failures, where an alcohol rehab page ranks for the pornographic query "Sister Maria treatment in the night by the aen work" despite containing zero target keywords.

The relevance engine erroneously bridges the semantic gap because the page mentions night (consumption context), sex (side effect of alcohol), and connects Mother Teresa (topically adjacent to Sister Maria), reveals Koray Tuğberk Gübür.

Furthermore, the SERP dynamically bifurcates based on the time of day: during US nighttime, Google ranks porn sites for the query; during daytime, the intent shifts, and it pivots to rehab sites.

This confirms that relevance scores are not static but are fluidly re-weighted by temporal user intent signals in real-time.

The phenomenon escalates via the Now Boost effect, where click data misattributes relevance to entirely unmentioned topics.

Because the rehab page received clicks from the "Sister Maria" porn query, Google inadvertently assigned it rankings for porn addiction—a term entirely absent from the page's DOM.

This proves that high-velocity click data can overwrite semantic gaps and propagate relevance to unrelated topics based solely on user behavior.