The problem with BLAST is that it will show all hits that are similar to your query. Often, the closest hits will not agree on the same specific taxa (e.g. hit different species. Often if the hits are equivalent, the results are arbitrarily sorted. So, the top items in the list are not necessarily the best hits. The classifiers will collapse the taxonomy to the lowest common ancestor if your query hits multiple reference sequences equally well. Especially, if the sequence is identical across different taxa.
Here are a few threads on this topic: