We analyse the impact of galaxy–halo misalignment on the ability of weak lensing studies to constrain the shape of dark matter haloes, using a combination of the Millennium dark matter N-body simulation and different semi-analytic galaxy formation models, as well as simpler Monte Carlo tests. Since the distribution of galaxy–halo alignments is not known in detail, we test various alignment models, together with different methods of determining the halo shape. In addition to alignment, we examine the interplay of halo mass and shape, and galaxy colour and morphology with the resulting stacked projected halo shape. We find that only in the case where significant numbers of galaxy and halo minor axes are parallel does the stacked, projected halo axis ratio fall below 0.95. When using broader misalignment distributions, such as those found in recent simulations of galaxy formation, the halo ellipticity signal is washed out and would be extremely difficult to measure observationally. It is important to note that the spread in stacked halo axis ratio due to theoretical unknowns (differences between semi-analytic models and between alignment models) are much bigger than any statistical uncertainty. It is naïve to assume that, simply because Λ cold dark matter (ΛCDM) predicts aspherical haloes, the stacked projected shape will be elliptical. In fact, there is no robust ΛCDM prediction yet for this procedure, and the interpretation of any such elliptical halo signal from lensing in terms of physical halo properties will be extremely difficult.