Satellite observations and reanalysis datasets are increasingly used to benchmark long-term variability in surface air temperature (SAT), an essential diagnostic for studying and quantifying changes in the Earth system. However, structural shifts in these time series—arising from both physical phenomena and operational artifacts—can introduce significant uncertainties. Furthermore, increasing interdependence between satellite retrieval algorithms and reanalysis products complicates the interpretation of SAT variability derived from these datasets. This study evaluates the consistency of global SAT records over the past two decades (2002-2023) by intercomparing major reanalysis products, operational NASA retrievals from the Atmospheric Infrared Sounder (AIRS), and gridded in-situ records. Using multiple linear regression on deseasonalized monthly time series, we identify discrepancies in trend magnitude and spatial distribution, particularly in data-sparse regions such as sub-Saharan Africa and the Southern Ocean. We then apply kernel changepoint detection (KernelCPD) with linear and Gaussian kernels to detect abrupt distributional shifts in the time series. Our analysis distinguishes physical changepoints common to all datasets, such as abrupt warming associated with the 2015–2016 El Niño, from artificial dataset-specific shifts. Notably, we identify synchronized changepoints in MERRA-2 and Community Long-term Infrared Microwave Combined Atmospheric Product System for Aqua (CLIMCAPS-Aqua) tied to sea ice forcing transitions in data assimilation and shifts in AIRS Version 7 retrievals coinciding with known instrument recalibrations. These findings display the impact of algorithmic dependence on data consistency, highlighting the need for further development of satellite retrieval frameworks, and demonstrate the value of changepoint detection as a diagnostic tool for validating long-term climate records.