New AI Makes Search Smarter by Asking You Questions
Imagine search engines that actually ask what you mean instead of guessing wrong.
Deep Dive
Deep search stumbles when queries are vague—missing time, location, or scope. A new benchmark, Clarify-Then-Search, asks whether AI-generated clarifying questions actually help. Built on 518 real Baidu queries, it measures end-to-end search quality against evidence-grounded nuggets. The verdict: asking one clarification question already beats no interaction, and bigger question budgets generally help more. But there's a catch—models often over-ask about region details that users can't answer, wasting the budget.
Key Points
- AI search engines often guess wrong when your question is vague—like 'best laptop' without saying for gaming or work.
- A new system tests if asking 1–3 clarifying questions first improves results (it does).
- Early tests on real search data show clearer answers, but some AIs still ask unanswerable questions.
Why It Matters
Could save you time and frustration every time you search online by making results actually match what you need.