There are many ways in which an alignment can look awful different types of "bad" each have their own different cause(s).  If you could upload a sample of what your bad alignment looks like, or the whole alignment, we could better give you help.

I went to RepBase and looked at the AluJb sequence:

http://www.girinst.org/protected/repbase_extract.php?access=AluJb

I BLASTED it against GenBank, saved the top 200 "hits" and aligned them.  Then I cleaned the alignment up a little bit, mainly deleting a few sequences which were far shorter than the rest.  And I used the SNAD, Sequence Name Annotation-based Designer to replace the BLAST output sequence names with accession_genus.species_position where position is the location of the AluJb repeat within the Accession# sequence which was often a whole chromosome sequence.

I am attaching the alignment, and a quick tree built from that alignment.  I am no expert on Alu repeats.   A couple of things seem odd to me:

1) I found hundreds of Homo sapiens, Pan troglodytes and Maccaca mulata Alu repeats with my BLAST search but not Gorilla gorilla, Pan paniscus, Orangutan, or other primate sequences.  I know there are genome projects for many more species in between human and Macaca mulata.  So I do not understand why my search did not pull them in.

2) A lot of human DNA is more than 98% identical to chimpanzee DNA, even in noncoding regions.  So I would have expected to find many human/chimp pairs very similar in my results.  Instead I am finding a wide distribution of sequences with human, chimp and macaque roughly equidistant to each other.  I don't know what to make of that, but my first guess is some oddity of my BLAST search rather than true biology.

http://veb.lumc.nl/SNAD/

A smaller alignment and tree built from it.  This one has a few from the RepBase and a few from primates.  I found many more Alu repeats in gorilla, Pan paniscus, orangutan, etc.  Not sure why they were missed in my first search.

I found that when I BLAST against GenBank, if I select one "hit" that is for example 97% identical to the query, it actually downloads dozens of other Alu repeats from that same genome entry.  So I am not sure how to get just the top matches, when searching with highly repetitive elements like Alu.

MAFFT with L-INs-i parameteres usually give better results than Muscle