No, it is not correct. At best you will have erroneous SNPs predicted all over your gene with very little support (low specificity) or way too high sensitivity and will lose a lot of SNPs that can be called using proper statistic methods. What you are explaining can be done with very degraded law coverage and quality ancient DNA where there are very few good variant and haplotype callers. For a high-quality variant calling, you either need a tool dedicated for your type of samples, sequencing equipment, and chemistry, or you may try to build your own statistical model, but that makes sense nowadays only for special cases. Otherwise, you are better with using multiple tools with different parameters and aggregating data in a certain way from them, if you do not want to rely on a single tool.