Maybe you just have a rooting problem? Try rerooting at some other node, maybe you OTU 1 and clone's sequence cluster then. Consider the tree as a network rather than a tree. If you know the basal taxa in your Alignment, try to set them as an outgroup and root the tree there during analysis (e.g. RAxML, Paup, etc)

That's a good suggestion and I have tried playing around with adding different outgroups to help root the tree different ways, but most of my sequences are novel on the Order/Family level, so it is tough and I keep getting a lot of paraphyly. However, part of the problem is that there is a lot of evolutionary distance between OTU 1 and the clone on the tree, which is just weirding me out since OTU 1 BLAST results comes up as the clone. So I am thinking something is going weird in the alignment (using muscle) that somehow changes it so that the clone and OTU 1 aren't actually closely related. I've just never seen something like this happen... 

That may just point to a lack of sequences with more identities than your clone/OTU1 in the database you blast compared to the sequences in your alignment. Additionaly, Blast works different than a sophisticated tree inference algorithm with substitution models, invariant sites and such. Do you partition the data, e.g. assign different models to 1., 2., 3. codon positions? Or do you infer the translated protein sequence, e.g. in likelihood or bayesian context?

This is where my knowledge of how the programs I use breaks down a bit, as I am not really sure what the methods I use do. I have been using two programs to try different ways to make trees, I have been using IQTREE, which I like because it tests different substitution models and runs a ML tree, it is easy to use and online (http://iqtree.cibiv.univie.ac.at/). I have also been using RAxML and specifying their GTRCAT model. Both giving similar topologies and bootstrap values. I have relatively few taxa that I am running (31), with short sequences (~420bp) and am definitely open to trying/learning different things. 

IQ-Tree is very interesting, since it can do free rate heterogeneity. This is cool in big data sets, but yours is rather small. Firstly, GTRCAT is an approximation to GTRGAMMA, the author recommends that for data sets >50 taxa. Better use GTRGAMMA via -m in raxml. Not shure if that helps on your data, but well...

blast results should resemble phylogenetic knowledge but it is not a tool to infer relationships.

Try the opposite: test the topology you expect (build it manually a tree editor) via approximately unbiased or shimodeira hasegawa test (consel) against your alignment, and if you have a significant p-value that your topology is wrong, you should question those blast results.

Thanks so much for your advice Michael! I have learned a few things from our discussion. But turns out the problem was my sequence data and Muscle. I don't know the particulars of how muscle's algorithms work, but once I switched to http://mafft.cbrc.jp/alignment/server/ for my alignments, and used mafft's ability to check directionality of my sequences, I was able to get much better alignments and now my OTUs are clustering with known taxa etc and my data is way more interesting now. 

Thanks again!