Original title: When Retrieval Helps: Selective Retrieval for Single-Turn Mental-Health QA
Hyunseo Oh, Chong-Kwon Kim, Yoonhyuk Choi · September 2026
When people ask AI about mental health, their questions often mix deep emotional distress with safety risks. Many systems use retrieval-augmented generation, where the AI pulls in outside documents before answering. However, searching for outside information is not always helpful. Researchers tested when pulling in external guidelines helps or hurts. They built a system that only retrieves outside text when an answer actually needs it. The system checks three things—coping needs, educational needs, and response detail—plus an automatic safety check. They found that always pulling in outside information adds detail, but it lowers overall answer quality and causes extra safety failures. Choosing selectively when to search protects answer quality and avoids the damage caused by searching every single time.
You're cooking a new dish and have the cookbook open. Constantly flipping back to the cookbook for every tiny detail slows you down and might make you add too much salt in the rush. In this paper, always using retrieval in mental-health chatbots is like constantly checking the cookbook, which can lead to worse answers. Selective retrieval checks only when needed, preserving the flow and safety.
For people in distress seeking mental-health advice, using retrieval in chatbots can sometimes lead to unsafe or less helpful responses. This matters because the right approach needs to balance helpful information with safety.
The study focuses on single-turn interactions and may not apply to multi-turn conversations. It uses specific datasets and a fine-tuned generator, so results might vary with other models or contexts. The retrieval need is operationalized with defined dimensions, which might not capture all nuances of mental-health queries.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
When Retrieval Helps: Selective Retrieval for Single-Turn Mental-Health QA — Hyunseo Oh, Chong-Kwon Kim, Yoonhyuk Choi
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.