Some suggestions that could be coded into search algorithms:

• put some categories for the type of information you want – e.g. news, sport, academic, media, hobbies, tech, shopping, and so on. Put a dropdown list by the criteria box to choose the type of sites to search.
• provide results that actually fit into the type of search request. Don’t make me go into detail
• If it’s an academic text, why not miss out all the links that are just book adverts or that tell you you can access the information if you have access to blah university’s intranet.
• miss out all the refs that are just code in someone’s database

OK I think Google et al do a fantastic job, given their limitations. However, more human input might make search results make more sense. If this would be too expensive to implement, why not have every site mark the areas that they wanted people to be able to find – not just keywords or descriptions and so on, all too easy to manipulate by experts – but some addition to the w3 standards that say just search the real content bits. If sites mess about by manipulating this, that would probably have to be by having content…..