Glossary · Technical
Model Weights
The billions of numerical parameters learned during training that encode everything a model knows and determine how it responds to a given input.
What they are
Training adjusts a model's weights repeatedly until its outputs match patterns in the training data as closely as possible. Once training ends, the weights are fixed — running the model on a new prompt (inference) doesn't change them.
Why it matters for AI visibility
A brand's presence baked into a model's weights — what it "remembers" from training — is a completely different thing from what it retrieves live during a conversation. Weights only update when a new model version is trained and released, so a brand's trained-in reputation can lag well behind its current reality.
See more terms in the full glossary, or read the AI Visibility 101 guide for the full picture.