See also
Quality control for OTU sequences
This problem applies to 97% OTU clustering, not to denoising (ZOTUs). The cluster_otus command is designed to generate a set of OTU sequences that are <97% identical to each other. If a pair of OTU sequences is >=97% identical, this is a mistake by the algorithm which is sometimes caused by the heuristics used for speed.
This is a pretty benign problem -- the 97% threshold is arbitrary anyway, so if some clusters are a bit "tighter", say 98% separated rather than 97%, it probably doesn't make any practical difference to the biological results.
To check for it, you can cluster with UCLUST with parameters chosen for high sensitivity rather than speed. For example,
usearch -cluster_fast otus.fa -id 0.97 -maxaccepts 4 -maxrejects 128 \
-top_hit_only -uc hits.uc -centroids new_otus.fa
With this command, there would ideally be no hits because each OTU should be in its own cluster. To check for hits and see the identities,
grep "^H" hits.uc | cut -f4 | sort -g
If you do get hits, then you can use new_otus.fa as an improved set of OTU sequences.