Glossary · Technical
Model Distillation
A training technique where a smaller "student" model is trained to mimic the outputs of a larger "teacher" model, producing a lighter, faster model with similar behavior.
What it is
Rather than training a small model from scratch on raw data, distillation trains it to reproduce the larger model's outputs, transferring much of the larger model's behavior into a fraction of the size and compute cost.
Why it matters for AI visibility
Many consumer-facing AI features — quick chat replies, mobile assistants, free-tier responses — run on distilled, smaller model versions for speed and cost reasons, and these can behave differently from their larger counterparts. That means a brand's visibility can genuinely vary depending on which model size actually answered a given query, even on the same platform.
See more terms in the full glossary, or read the AI Visibility 101 guide for the full picture.