How Specific Do I Need to Be? Effects of User Query Underspecification on Deep Research Agents
Deep research agents can stay grounded in cited evidence while missing what the user actually needs.
The researchers introduce UNDERSPEC-DR, a framework for testing how incomplete queries affect answer quality and search behavior. Across three agents, underspecification substantially reduced how well answers met users’ information needs. Search analysis suggests a possible explanation: agents searched more broadly, drawing less on the query and more on learned priors that could diverge from user intent. The findings point to a need for agents to recognize when they require more user input.
ArXiv · AI/CL/LG's note
The researchers introduce UNDERSPEC-DR, a framework for testing how incomplete queries affect answer quality and search behavior. Across three agents, underspecification substantially reduced how well answers met users’ information needs. Search analysis suggests a possible explanation: agents searched more broadly, drawing less on the query and more on learned priors that could diverge from user intent. The findings point to a need for agents to recognize when they require more user input.
ArXiv · AI/CL/LG's note
score 5