Large time-domain surveys provide a unique opportunity to detect and explore variability of millions of sources on timescales from days to years. Broadband photometric variability can be used as the key selection criteria for weak type-I active galactic nuclei (AGN), when other "direct" confirmation criteria like X-ray or radio emission are unavailable. However, to detect variability of rather weak AGN powered by intermediate-mass black holes, typical sensitivity provided by existing light curve databases is insufficient. Here we present an algorithm for post-processing of light curves for sources with stochastic variability, retrieved from the The Zwicky Transient Facility (ZTF) Forced Photometry service. Using our approach, we can filter out spurious data points related to data reduction artefacts and also eliminate long-term trends related to imperfect photometric calibration. We can now confidently detect the broadband variability at the 1-3 % level which can potentially be used as a substitute for expensive X-ray follow-up observations.