This paper presents a novel semantic overlay scheme that connects peers based on the similarity of their contents. Semantic closeness among overlay peers can effectively be determined via the exchanges of their content summary using Bloom filters. The overlay link quality is further improved by carefully selecting semantic neighbors according to their potential to contribute to content-based searches. The basic idea behind the semantic neighbor selection is that highly replicated documents should not excessively dominate the overlay topology, overshadowing rare to modestly-replicated items whose query efficiency is often more critical for overall search performance. The efficacy of the proposed semantic overlay is validated through our simulation study which demonstrates superior overlay link quality and query routing performance.