This sounds like an issue with your reference database (incomplete? poor specificity?), the gene itself (lack of taxonomic resolution?), or the primers (poor specificity?). You should go back and check all three to troubleshoot. Provided the reference database fully covers your amplicons, the vsearch-based classifier in q2-feature-classifier should work well, so we can rule out the classifier.
One thing is certainly true: many of the unhelpful classifications you are getting are actually how the reference sequences in your database are annotated, so it is clear that at least part of the issue is your database (missing annotations).
I am unfamiliar with this functional gene, and it sounds like you are using a custom database, so unfortunately I cannot offer any more insight, only this:
This is one reason why 16S is so popular: because it has been historically popular, and so the reference databases and protocols are well designed and validated, and tutorials and support can be found quite easily from others. When you start using other targets, you are entering mostly uncharted territory.