Driven by promising recent results, there has been a revived interest in the thermoelectric properties of organic (semi)conductors. Concomitantly, there is a need to probe the Seebeck coefficient S of modestly conducting materials in thin film geometry. Here we show that geometries that seem desirable from a signal-to-noise perspective may induce systematic errors in the measured value of S, Sm, by a factor 3 or more. The enhancement of Sm by the device geometry is related to competing conduction paths outside the region between the electrodes. We derive a universal scaling curve that allows correcting for this and show that structuring the semiconductor is not needed for the optimal electrode configuration, being a set of narrow, parallel strips.