Multimessenger observations of the neutron star merger GW170817 and its kilonova proved that neutron star mergers can synthesize large quantities of r-process elements. If neutron star mergers in fact dominate all r-process element production, then the distribution of kilonova ejecta compositions should match the distribution of r-process abundance patterns observed in stars. The lanthanide fraction (X La ) is a measurable quantity in both kilonovae and metalpoor stars, but it has not previously been explicitly calculated for stars. Here, we compute the lanthanide fraction distribution of metal-poor stars ([Fe/H] < −2.5) to enable comparison to current and future kilonovae. The full distribution peaks at log X La ∼ −1.8, but r-process-enhanced stars ([Eu/Fe] > 0.7) have distinctly higher lanthanide fractions; log X La −1.5. We review observations of GW170817 and find general consensus that the total log X La = −2.2 ± 0.5, somewhat lower than the typical metal-poor star and inconsistent with the most highly r-enhanced stars. For neutron star mergers to remain viable as the dominant r-process site, future kilonova observations should be preferentially lanthanide-rich (including a population of ∼ 10% with log X La > −1.5). These high-X La kilonovae may be fainter and more rapidly evolving than GW170817, posing a challenge for discovery and follow-up observations. Both optical and (mid-)infrared observations will be required to robustly constrain kilonova lanthanide fractions. If such high-X La kilonovae are not found in the next few years, that likely implies that the stars with the highest r-process enhancements have a different origin for their r-process elements.