Gallium nitride (GaN), a direct-gap semiconductor that is radioluminescent, can be used as a transducer yielding a high signal from a small detecting volume and thus potentially suitable for use in small fields and for high dose gradients. A common drawback of semiconductor dosimeters with effective atomic numbers higher than soft tissues is that their responses depend on the presence of low energy photons for which the photoelectric cross section varies strongly with atomic number, which may affect the accuracy of dosimetric measurements. To tackle this 'over-response' issue, we propose a model for GaN-based dosimetry with readout correction. The local photon spectrum is calculated by convolving fluence pencil kernel spectra with the beam aperture fluence distribution. The response of a GaN detector is modelled by combining large cavity theory and small cavity theory for the low and high energy components of the local spectrum. Monte Carlo simulations are employed for determination of specific correction factors for different GaN transducer sizes and irradiation conditions. Some model parameters such as the cut-off energy and partitioning energy are discussed. The accuracy of the GaN dosimetric response model has been evaluated for tissue phantom ratio experiments along the central axis. These experiments have shown that calculated and measured GaN responses stay within ±3% at all depths beyond the build-up depth. The calculated GaN response factor is also in good agreement with measured data (±2.5%). The validated model with response compensation improves significantly the accuracy of dosimetric measurements: below 2.5% deviation as compared to 13% without compensation, for a 10 × 10 cm(2) field, at depth from 1.5 to 22 cm.