Abstract The treatment of crowded fields in Gaia data will only be a reality in a few years from now. In particular, for globular clusters, only the end-of-mission data (public in 2022–2023) will have the necessary full crowding treatment and will reach sufficient quality for the faintest stars. As a consequence, the work on the deblending and decontamination pipelines is still ongoing. We describe the present status of the pipelines for different Gaia instruments, and we model the end-of-mission crowding errors on the basis of available information. We then apply the nominal post-launch Gaia performances, appropriately worsened by the estimated crowding errors, to a set of 18 simulated globular clusters with different concentration, distance and field contamination. We conclude that there will be 103–104 stars with astrometric performances virtually untouched by crowding (contaminated by <1 mmag) in the majority of clusters. The most limiting factor will be field crowding, not cluster crowding: the most contaminated clusters will only contain 10–100 clean stars. We also conclude that (i) the systemic proper motions and parallaxes will be determined to 1 per cent or better up to ≃15 kpc, and the nearby clusters will have radial velocities to a few km s−1; (ii) internal kinematics will be of unprecedented quality, cluster masses will be determined to ≃10 per cent up to 15 kpc and beyond, and it will be possible to identify differences of a few km s−1 or less in the kinematics (if any) of cluster sub-populations up to 10 kpc and beyond; (iii) the brightest stars (V ≃ 17 mag) will have space-quality, wide-field photometry (mmag errors), and all Gaia photometry will have 1–3 per cent errors on the absolute photometric calibration.