The Neuron: AI Explained AI vs Google Search....behind the scenes
Oct 10, 2025
Mark Williams-Cook, Director at Candour and founder of AlsoAsked, dives into the seismic shifts AI is causing in search dynamics. He labels AI a 'leaky bucket,' highlighting how misinformation and spam are rampant. Discover how expired domains manipulate language models and what the collapse of the link graph means for trust in search. Mark shares strategies for improving AI visibility and discusses the risks of brand impersonation. He also predicts that authentic human content will see a resurgence as users seek credibility.
AI Snips
Chapters
Books
Transcript
Episode notes
LLM Non-Determinism Scales Badly
- Large language model outputs are non-deterministic and small error rates scale to massive harm at web scale.
- Mark warns that occasional wrong answers get reinforced by user feedback loops and spread rapidly.
Candor Reviews Were Wrongly Merged
- Mark described how Google/AI conflated reviews for two companies both named Candor, which produced incorrect pros/cons.
- That example shows how grounding failures can damage real reputations and mislead users.
Schema Links Don't Survive Training
- Tokenization in model training strips structured schema connections so entity links often vanish in LLMs.
- Mark explains that this loss of explicit structure causes models to confuse entities and relationships.
