Incomplete RAG answers may be worse than obvious false answers
RAG systems answer by first pulling in documents, but they may not retrieve all the documents needed for a full answer. The result can look reliable because it has and every sentence may be supported by the documents that were retrieved.
The deeper problem is that important evidence may be missing from the , so the final conclusion can still be wrong. This is harder to spot than because nothing obviously fabricated appears in the answer.
Users may trust the , while the system has no clear way to warn that it only saw part of the available evidence. In real products, completeness may matter more than how well the model writes the final response.
Key points
- A RAG answer can be fully cited and still miss key evidence.
- Incomplete can make the overall conclusion wrong even when each sentence is supported.
- This failure is harder to detect than because the answer looks grounded.
- systems may need to focus more on completeness than writing quality.
- Cutting token use too aggressively can raise the risk of incomplete answers.