Machine-learned language understanding in ranking: RankBrain (confirmed October 2015), BERT (October 2019, ~10% of English queries), MUM (announced 2021)

Claim:

• RankBrain — machine-learning ranking component, publicly confirmed October 2015 (live since ~early 2015). Google called it one of its top-three ranking signals at the time. Helps interpret never-before-seen queries.
• BERT — transformer-based natural-language understanding, October 2019, initially affecting ~10% of English queries.
• MUM — announced 2021.

Source: Google + Bloomberg (RankBrain, October 2015); Google blog (BERT, October 2019); Google I/O 2021 (MUM).

Confidence: Verified (dates + descriptions); Industry-consensus on relative importance.

Why this matters for Candid: Reinforces the Hummingbird lesson — write naturally, answer real intent. You cannot directly "optimise for BERT"; you optimise by being clear and genuinely useful. Strengthens the case for Rule: Restructure top service and FAQ pages to answer the customer's question directly in the first 100–150 words, with specific local detail.