Skip to main content
Core concept · AI Search Infrastructure

​
Definition

Knowledge conflict is the condition that occurs when a model’s parametric knowledge contradicts retrieved content. Resolution is not deterministic — sometimes retrieval wins, sometimes parametric memory wins, sometimes the model hedges by presenting both. Research consistently documents this as producing irrational and inconsistent behavior. Knowledge conflict is the mechanism behind the parametric inertia problem. When a brand has repositioned, the conflict between the model’s trained belief and the current retrieved evidence is real — and the model does not resolve it reliably. The practical implication: changing what a model says about an established brand requires changing the parametric layer, not just optimizing retrievable content.

Parametric inertia

Parametric knowledge

Post-hoc citation

Training cutoff

RAG

​
Relevant Plate Lunch Collective Services

AI SEO Context Map AI Search Visibility Assessment