See also
OTU / denoising pipeline
Read preparation
Paired read merging
Edgar & Flyvbjerg 2014 paper on expected errors and
Bayesian assembly
If you have paired reads, pick a couple of samples and try running the fastq_mergepairs command. Verify that most of the pairs align and that the lengths of the consensus sequences are consistent with the distribution expected for your primer pair. See links from the fastq_mergepairs command page for more documentation about adjusting parameters, verifying the results and trouble-shooting problems.
Paired read merging should be performed before quality filtering because merging improves base call error estimates (Q scores) in the overlapping segment.
Example, 2x250 V4
With very long overlaps such as typical 2x250 V4 you will sometimes find that it is better to allow more differences in the alignment compared to the default parameters; see example report. For example, you might choose to increase set -fastq_maxdiffs 10 -fastq_pctid 80:
usearch -fastq_mergepairs *R1*.fastq -relabel @ -fastq_maxdiffs 10
\
-fastq_pctid 80 -fastqout merged.fq
This command uses a couple of convenient features supported by fastq_mergepairs: merging several samples into one file by using shell wildcards (*R1*), and automatically embedding the sample name into the read label by using -relabel @.