RAG & Retrieval · Ground answers in evidence
How is RAG different from building a search engine?
Short answer
Search returns ranked documents for a human to read. RAG retrieves evidence and asks a model to synthesise an answer. RAG inherits search problems and adds generation risk.
A search engine optimises for finding documents; the user interprets them. RAG optimises for answering a question; the model interprets the documents. That means RAG needs everything search needs, plus faithfulness, citation and answer-level evaluation.
- Search relevance is evaluated with ranking metrics; RAG adds answer correctness and groundedness.
- Search users can scan several results; a RAG answer collapses them into one claim set.
- Search failure is visible; RAG failure can sound authoritative.
- Query understanding, indexing and ranking remain necessary underneath RAG.
Follow-ups you should expect
Interviewers use these to check whether the first answer was understanding or memorisation.
- 1Which search metrics still matter in RAG?
- 2How would you show sources alongside a generated answer?
- 3When would you ship search instead of RAG?
Say it in the interview
Search returns documents and lets the human do the synthesis. RAG takes the next step and generates the answer. That means RAG still needs solid indexing, query understanding and ranking; it just adds generation on top. The risk is that a retrieval failure now looks like a confident answer instead of an empty result. So I keep search-quality metrics and add groundedness and citation checks.
Use this as a model, not a script. Saying it in your own words is what makes it survive follow-up pressure.
Related questions
Reading is step one
Now say it out loud.
Understanding an answer and delivering it under pressure are different skills. Rehearse this topic in a live RAG & Retrieval mock interview and get scored feedback on depth, clarity and confidence.