Sir

It is disturbing that no single search engine indexes more than a meagre 16 per cent of the web, down from about 33 per cent 18 months ago ( Nature 400, 107–109; 1999). Users' ability to look at documents has remained static, so people still only look at the first few tens of search results.

Most mainstream search engines rely on relevance-score algorithms to rank matches. Such algorithms are prone to manipulation by content providers, who can rig their content to yield a spuriously high relevance score. But transparent algorithms are necessary for users to understand how to get the most from search engines. The end result is a long list of hits of doubtful relevance, quality and completeness.

This makes topic-specific search engines — which aim to cover the majority of content within a specific topic, rather than a small fraction of every topic — all the more appealing. Such an approach also allows content to be reviewed and structured so that users are presented with an intelligently categorized, hierarchical list of matches, rather than a linear one. This allows rapid identification (‘drilling down’) of the most relevant matches.

One example that addresses these issues is OrthoSearch (http://www.orthosearch.com ), covering orthopaedics and related topics. OrthoSearch uses the Internet Society of Orthopaedic Surgery's web-links policy to determine which sites are relevant, avoiding many shortcomings of the most popular search engines.