Another take: in most circumstances, particularly when the task is not a known-item query, searchers don't know the precise keywords they need to type in order to acquire their target. So although the mechanism or function they are using is a "search", actually they want to explore and browse the information space. Jared Spool and UIE have explored this topic
Filtering - if we are talking about iterative, faceted search - allows users to enter a vague term and then iteratively refine it. This is the "Paradox of the active user" in action.
These "active users" don't have time to learn about the information space so they go ahead and search anyway. When the results are received for that search, they then start learning about the information space and refine/restart their query.
The design principle of timely feedback works all the way through this experience, so "Spotlight-style" results all go towards helping make the query reflective of the information space and of the person's original intention.
For a good example of both these techniques at work, try to find yourself a new TV at John Lewis or any number of e-commerce sites.