Context. Over the past few years, new multiplex spectrographs have emerged to observe several millions of stars. The optimisation of these instruments (w.r.t. their resolution or wavelength range), their associated surveys (choice of instrumental set-up), and their parameterisation pipelines require methods that estimate which wavelengths (or pixels) contain useful information. Aims. We propose a method that establishes the usefulness of an atomic spectral line, whereby usefulness is defined by the purity of the line and its detectability. We demonstrate two applications of our code: a) optimising an instrument by comparing the number of detected useful lines at a given wavelength range and resolution; and b) optimising the line list for a given set-up, in the sense of creating a golden subsample of the least-blended lines that are detectable at a range of signal-to-noise ratio values. Methods. The method compares pre-computed normalised synthetic stellar spectra containing all of the elements and molecules with spectra solely containing the lines of specific elements. Then, the flux ratios between the full spectrum and the element spectrum are computed to estimate the line purities. The method automatically identifies: (i) the line's central wavelength, (ii) its detectability based on its depth and a given signal-to-noise threshold, and (iii) its usefulness based on the purity ratio defined above. Results. We applied this method to compare the three WEAVE high-resolution set-ups (blue: 404-465 nm, green: 473-545 nm, red: 595-685 nm) and find that the green+red set-up both allows us to measure more elements and contains more numerous useful lines. However, there is a disparity in terms of which elements are detected over each of the set-ups that we have characterised. We also studied the performances of high-resolution (R similar to 20 000) and low-resolution (R similar to 6000) spectra covering the entire optical wavelength range. Assuming a purity threshold of 60%, we find that the high-resolution set-up contains a much wealthier selection of lines, for any of the considered elements; whereas the low-resolution set-up displays a 'loss' of 50% to 90% of the lines (depending on the nucleosynthetic channel considered), even when the signal-to-noise ratio is increased. Conclusions. The method presented here provides a vital diagnostic of where to focus to get the most out of a spectrograph. It is easy to implement for future instruments that have not yet determined their final configuration, as well as for pipelines that require line masks.