Skip to main content
ModelTerms

Comparison

Hallucination vs Retrieval-Augmented Generation

Hallucination and Retrieval-Augmented Generation are both common AI/LLM terms but cover different ideas. Here is a quick side-by-side.

When you would reach for Hallucination

Hallucination comes up when the question is fundamentally about evaluation.

Citing a paper that does not exist.

When you would reach for Retrieval-Augmented Generation

When the model needs information that is not baked into its weights — fresh, private, or domain-specific.

"Chat with your PDFs" — Notion, Glean, ChatGPT custom GPTs.

Frequently asked

What is the difference between Hallucination and Retrieval-Augmented Generation?

Hallucination: A hallucination is a confidently-stated, plausible-sounding LLM output that is factually wrong. It is the failure mode that most often surprises non-expert users. Retrieval-Augmented Generation: RAG retrieves relevant documents from a corpus at query time and includes them in the prompt, letting an LLM answer with up-to-date, source-cited, private information without retraining.

When should I use Hallucination vs Retrieval-Augmented Generation?

Hallucination is the right concept when you are focused on evaluation. When the model needs information that is not baked into its weights — fresh, private, or domain-specific.

Are Hallucination and Retrieval-Augmented Generation the same thing?

No. Hallucination is evaluation; Retrieval-Augmented Generation is agents & tools. They are related but address different parts of the AI stack.