Hi Gal,
See here. Kingdom-level classification usually indicates:
- the wrong database was used
- the wrong primers were used for extraction or the wrong type of data was input to the classifier (e.g., training a V4 classifier but attempting to classify V3-V4)
- inappropriate trim lengths are being used in
extract-reads. This should either match the length of your sequences, or just don't use it at all (e.g., if you have overlapping paired-end reads)
Mixed classification (e.g., lots of kingdom-level but also deeper classifications) is a more complicated problem that can indicate user error or data issues, but kingdom-only is almost always user errors (and it happens to all of us 
)
Please double-check your inputs and parameters to make sure that the primers, trim length, and database are all appropriate for the input data.
Please let us know if you are still having issues!