Cosmological simulations predict that galaxies are embedded into triaxial dark matter haloes, which appear approximately elliptical in projection. Weak gravitational lensing allows us to constrain these halo shapes and thereby test the nature of dark matter. Weak lensing has already provided robust detections of the signature of halo flattening at the mass scales of groups and clusters, whereas results for galaxies have been somewhat inconclusive. Here we combine data from five weak lensing surveys (NGVSLenS, KiDS/KV450, CFHTLenS, CS82, and RCSLenS, listed in order of most to least constraining) in order to tighten observational constraints on galaxy-scale halo ellipticity for photometrically selected lens samples. We constrain fh, the average ratio between the aligned component of the halo ellipticity and the ellipticity of the light distribution, finding fh = 0.303−0.079+0.080 for red lens galaxies and fh = 0.217−0.159+0.160 for blue lens galaxies when assuming elliptical Navarro-Frenk-White density profiles and a linear scaling between halo ellipticity and galaxy ellipticity. Our constraints for red galaxies constitute the currently most significant (3.8σ) systematics-corrected detection of the signature of halo flattening at the mass scale of galaxies. Our results are in good agreement with expectations from the Millennium Simulation that apply the same analysis scheme and incorporate models for galaxy–halo misalignment. Assuming these misalignment models and the analysis assumptions stated above are correct, our measurements imply an average dark matter halo ellipticity for the studied red galaxy samples of ⟨|ϵh|⟩ = 0.174 ± 0.046, where |ϵh| = (1 − q)/(1 + q) relates to the ratio q = b/a of the minor and major axes of the projected mass distribution. Similar measurements based on larger upcoming weak lensing data sets can help to calibrate models for intrinsic galaxy alignments, which constitute an important source of systematic uncertainty in cosmological weak lensing studies.