Skip to content
VectorCite
Free audit

Glossary entry fine-tuning

Fine-tuning

When a model is further trained on a specific corpus. It's distinct from RAG (which uses retrieval at inference time, not training).

Filed under
Core concepts

Definition

Fine-tuning is the process of further training a foundation model on a domain-specific corpus to specialize its outputs. Distinct from RAG: fine-tuning bakes knowledge into the model's weights; RAG injects it at query time via retrieval. Most answer engines do NOT cite from fine-tuned knowledge. They cite from RAG-retrieved live web content. So AEO is about being in the retrieval corpus, not the fine-tune corpus.

Example

An OpenAI fine-tune for legal queries has lawyer-style outputs but still cites the LIVE web via tool use for current case law. The cited sources are RAG results, not fine-tune data.

How to optimize

Don't worry about being in fine-tune corpora. They're rarely updated and cite-free. Focus on being retrievable: indexed, schema-marked, BLUF, dated.

Check a page against a real AI answer

Claude’s live answer to your buyer’s question, its sources, your readiness score and the fixes. Free, no signup, about half a minute.

Run a free audit