The detection of new clusters of galaxies or the study of known clusters of galaxies in X-rays can be complicated by the presence of X-ray point sources, the majority of which will be active galactic nuclei (AGN). This can be addressed by combining observations from a high angular resolution observatory (such as Chandra) with deeper data from a more sensitive observatory that may not be able to resolve the AGN (like XMM). However, this approach is undermined if the AGN varies in flux between the epochs of the observations. To address this we measure the characteristic X-ray variability of serendipitously detected AGN in 70 pairs of Chandra observations, separated by intervals of between one month and thirteen years. After quality cuts, the full sample consists of 1511 sources, although the main analysis uses a subset of 416 sources selected on the geometric mean of their flux in the pairs of observations, which eliminates selection biases. We find a fractional variability that increases with increasing interval between observations, from about 0.25 for observations separated by tens of days up to about 0.45 for observations separated by ∼10 years. As a rule of thumb, given the precise X-ray flux of a typical AGN at one epoch, its flux at a second epoch some years earlier or later can be predicted with a precision of about 60% due to its variability (ignoring any statistical noise). This is larger than the characteristic variability of the population by a factor of √2 due to the uncertainty on the mean flux of the AGN due to a single prior measurement. The precision can thus be improved with multiple prior flux measurements (reducing the √2 factor), or by reducing the interval between observations to reduce the characteristic variability.