Semantic search needs selectivity to improve its effectiveness. Traditional vector similarity search is limited by the need for exact predicates, whereas pre-filtering allows users to specify filter queries as part of the kNN attribute in the query, restricting the documents over which a kNN search will be performed. This enables users to limit their search to specific locations or criteria, such as city fields, and improves the accuracy of search results. Pre-filtering works by first filtering out documents that do not match the specified filter queries at a segment level, and then performing a kNN search on the remaining eligible documents. This approach allows for more efficient searching and better recall rates, making it an important development in semantic search technology.