Errors in multiple sequence alignments (MSAs) can reduce accuracy in positive-selection inference. Therefore, it has been suggested to filter MSAs before conducting further analyses. One widely used filter, Guidance, allows users to remove MSA positions aligned with low confidence. However, Guidance’s utility in positive-selection inference has been disputed in the literature. We have conducted an extensive simulation-based study to characterize fully how Guidance impacts positive-selection inference, specifically for protein-coding sequences of realistic divergence levels. We also investigated whether novel scoring algorithms, which phylogenetically corrected confidence scores, and a new gap-penalization score- normalization scheme improved Guidance’s performance. We found that no filter, including original Guidance, consis- tently benefitted positive-selection inferences. Moreover, all improvements detected were exceedingly minimal, and in certain circumstances, Guidance-based filters worsened inferences.