New ARSD metric quantifies search work compressed by AI answers
A single AI answer may replace dozens of web queries—now there's a way to measure that.
Conversational AI systems like ChatGPT and Perplexity can collapse a sequence of web queries, result inspections, and source comparisons into one synthesized answer. But until now, there was no standard way to measure how much conventional search work that answer actually represents. Existing retrieval metrics focus on ranking quality, user effort, or factual correctness—none capture the compression ratio between traditional browsing and a single AI reply.
Benjamin Tannenbaum's new paper, submitted to arXiv in July 2026, defines Answer-Reconstruction Search Density (ARSD): the minimum number of distinct query actions required, under a fixed policy, to support a target share of atomic retrievable answer units. A parallel page-density measure isolates compression from queries versus compression from sources. This gives researchers and developers a quantitative tool to compare how much 'hidden search' different conversational models perform, and to optimize systems for efficiency and transparency.
- ARSD measures the minimum distinct queries needed to reconstruct a conversational answer's atomic units.
- A parallel page-density metric separates query compression from source compression.
- Fills a gap in retrieval evaluation by quantifying the traditional search work replaced by AI answers.
Why It Matters
Provides a concrete metric to audit AI efficiency and compare conversational search against traditional web searching.