Background The limited variation observed among SARS-CoV-2 consensus sequences makes it difficult to reconstruct transmission linkages in outbreak settings. Previous studies have recovered variation within individual SARS-CoV-2 infections but have not yet measured the informativeness of within-host variation for transmission inference. Methods We performed tiled amplicon sequencing on 307 SARS-CoV-2 samples, including 130 samples from 32 individuals in 14 households and 47 longitudinally sampled individuals, from four prospective studies with household membership data, a proxy for transmission linkage. Results Consensus sequences from households had limited diversity (mean pairwise distance, 3.06 SNPs; range, 0-40). Most (83.1%, 255/307) samples harbored at least one intrahost single nucleotide variant (iSNV; median: 117; IQR: 17-208), above a minor allele frequency threshold of 0.2%. Pairs in the same household shared significantly more iSNVs (mean: 1.20 iSNVs; 95% CI: 1.02-1.39) than did pairs in different households infected with the same viral clade (mean: 0.31 iSNVs; 95% CI: 0.28-0.34), a signal that decreases with increasingly stringent minor allele frequency thresholds. The number of shared iSNVs was significantly associated with an increased odds of household membership (aOR: 1.35; 95% CI: 1.23 -1.49). However, the poor concordance of iSNVs detected across sequencing replicates (24.8% and 35.0% above a 0.2% and 1% threshold), confirms technical concerns that current sequencing and bioinformatic workflows do not consistently recover low frequency within-host variants. Conclusions Shared within-host variation may augment the information in consensus sequences for predicting transmission linkages. Improving sensitivity and specificity of within-host variant identification will improve the informativeness of within-host variation.