IP geolocation - the process of mapping network identifiers to physical locations - has myriad applications. We examine a large collection of snapshots from a popular geolocation database and take a first look at its longitudinal properties. We define metrics of IP geo-persistence, prevalence, coverage, and movement, and analyse 10 years of geolocation data at different location granularities. Across different classes of IP addresses, we find that significant location differences can exist even between successive instances of the database - a previously underappreciated source of potential error when using geolocation data: 47% of end users IP addresses move by more than 40 km in 2019. To assess the sensitivity of research results to the instance of the geo database, we reproduce prior research that depended on geolocation lookups. In this case study, which analyses geolocation database performance on routers, we demonstrate impact of these temporal effects: median distance from ground truth shifted from 167 km to 40 km when using a two months apart snapshot. Based on our findings, we make recommendations for best practices when using geolocation databases in order to best encourage reproducibility and sound measurement.