Sea Surface Temperature (SST) is an essential variable for understanding key physical and biological processes. Blended and interpolated L4 SST products offer major advantages over alternative SST data sources due to their spatial and temporal completeness, yet their ability to discriminate upwelling-induced steep temperature transitions in coastal waters remains largely unassessed. Here we analysed the performance of eleven L4 GHRSST-compliant products in estimating in situ water temperatures recorded by a large network of shallow subtidal and intertidal temperature loggers deployed in shores covering regimes with a wide range of upwelling intensities. Results indicate that while most products perform satisfactorily for most of the year, performance is severely affected during the upwelling season in locations with strong upwelling. We show that upwelling negatively impacts all four metrics used to assess dataset performance (average bias, correlation, centred root-mean-square error and normalized standard deviation), leading to a considerable overestimation of coastal water temperatures (with average bias exceeding 2 °C in some cases). We also show that while the use of L3 data (i. e., prior to blending and interpolation) leads to an increase in performance compared to L4 GHRSST-compliant products, the gain is probably not substantial enough to offset issues related with their spatial and temporal inconsistency along coastlines. Our results suggest that the use of L4 GHRSST-compliant products can lead to a misrepresentation of the thermal fingerprint of upwelling, and thus should be limited (or even avoided) in locations dominated by its effects. Conversely, the use of L4 GHRSST-compliant products on locations with little to no upwelling appears to be warranted. The mismatch between in situ and remotely-sensed sea water temperatures here reported also highlights the need for implementation of long-term monitoring networks of in situ temperature loggers.