One of the most peculiar characteristics of active galactic nuclei (AGNs) is their variability over all wavelengths. This property has been used in the past to select AGN samples and is foreseen to be one of the detection techniques applied in future multi-epoch surveys, complementing photometric and spectroscopic methods. In this paper, we aim to construct and characterise an AGN sample using a multi-epoch dataset in the r band from the SUDARE-VOICE survey. Our work makes use of the VST monitoring programme of an area surrounding the Chandra Deep Field South to select variable sources. We use data spanning a six-month period over an area of 2 square degrees, to identify AGN based on their photometric variability. The selected sample includes 175 AGN candidates with magnitude r<23mag. We distinguish different classes of variable sources through their lightcurves, as well as X-ray, spectroscopic, SED, optical, and IR information overlapping with our survey. We find that 12% of the sample (21/175) is represented by supernovae (SN). Of the remaining sources, 4% (6/154) are stars, while 66% (102/154) are likely AGNs based on the available diagnostics. We estimate an upper limit to the contamination of the variability selected AGN sample =~34%, but we point out that restricting the analysis to the sources with available multi-wavelength ancillary information, the purity of our sample is close to 80% (102 AGN out of 128 non-SN sources with multi-wavelength diagnostics). Our work thus confirms the efficiency of the variability selection method, in agreement with our previous work on the COSMOS field. In addition we show that the variability approach is roughly consistent with the infrared selection.