Researchers from Tencent Hunyuan and Tsinghua University found that AI search agents struggle with ambiguous queries, not the search itself. They created a benchmark called DiscoBench to test language models’ ability to spot ambiguity and ask follow-up questions. The study revealed that even large models have low accuracy rates, with the highest being 43.1 percent. This highlights the importance of developing AI that can effectively handle unclear or incomplete queries. Why it matters: improving AI’s ability to ask clarifying questions is crucial for enhancing its overall performance and reliability in real-world applications.