Significance The success of cancer treatment largely depends on the genetic mutations present within metastases, which cause 90% of cancer-related deaths. Genetically diverse metastases are more likely to harbor resistance mutations, contributing to treatment failure. It is often assumed that each metastasis is seeded exactly once, such that its diversity cannot be inherited and instead must emerge entirely during growth, yet many metastases have a diversity pattern inconsistent with this assumption. We introduce a mathematical model of consecutive seeding by multiple cells that can explain these patterns. We then apply this model to tumor sequencing data to infer that 10 to 150 cells seeded each metastasis. We derive predictions for the fraction of transferred diversity and the proportion of polyclonal lesions.