Did you know Google "re-ranks" search results using AI that is trained on user data and quality rater scores? In this video, I break down revelations from the DOJ trial, including the "Glue" system, RankEmbed BERT, and why user satisfaction is now the most critical factor for SEO. 00:00 - Intro: Why user satisfaction is the most important ranking factor. 00:49 - Google’s Quality Raters Guidelines & Helpful Content documentation. 01:36 - Marie’s Theory: Traditional algorithms vs. AI re-ranking systems. 02:35 - Why Google says "Don’t chunk your content" (Danny Sullivan). 03:00 - Vector Search & Embeddings: How AI retrieval actually works. 05:40 - The DOJ vs. Google Trial: Leaks on user-side data. 07:44 - The "Glue" System (NavBoost): How Google tracks clicks & hovers. 09:03 - "Instant Glue": How search adapts to new intent (Nice vs. nice). 10:20 - RankEmbed BERT: The deep learning model Google hid. 12:21 - How AI re-ranks the top search results. 13:03 - BERT Training: Wikipedia, Books Corpus... and romance novels? 15:18 - The 3 Deep Learning Models: RankBrain, DeepRank, RankEmbed. 17:40 - RankBrain: Why it only re-ranks the top 20-30 results. 18:47 - The Helpful Content System: A machine learning classifier. 21:18 - March 2024 Core Update: The evolution of helpfulness signals. 23:00 - Frozen vs. Retrained Google: How algorithms are tested. 26:01 - Why Google runs 17,000+ live traffic experiments. 27:16 - Does Google use Chrome data for rankings? (DOJ evidence). 28:32 - The "AI Mountain": Why AI-generated traffic often crashes. 29:30 - How to optimize for user satisfaction (beyond keywords). 31:00 - Liz Reid on AI Overviews: Why users click for depth.

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.

Transcript

Related