LISA will extend the search for gravitational waves (GWs) at $0.1\,{-}\,100$ mHz where loud signals from coalescing binary black holes of $ 10^4 \,{-}\,10^7\,\rm M_{\odot}$ are expected. Depending on their mass and luminosity distance, the uncertainty in the LISA sky-localization decreases from hundreds of deg$^2$ during the inspiral phase to fractions of a deg$^2$ after the merger. By using the semi-analytical model L-Galaxies applied to the Millennium-I merger trees, we generate a simulated Universe to identify the hosts of $z\,{\leq}\,3$ coalescing binaries with total mass of $3\,{\times}\,10^{5}$, $3\,{\times}\,10^6$ and $3\,{\times}\,10^7\rm M_{\odot}$, and varying mass ratio. We find that, even at the time of merger, the number of galaxies around the LISA sources is too large (${\gtrsim}\,10^2$) to allow direct host identification. However, if an X-ray counterpart is associated to the GW sources at $z\,{<}\,1$, all LISA fields at merger are populated by ${\lesssim}\,10$ AGNs emitting above ${\sim}\, 10^{-17} \, \rm erg\,cm^{-2}\,s^{-1}$. For sources at higher redshifts, the poorer sky-localization causes this number to increase up to ${\sim}\, 10^3$. Archival data from eRosita will allow discarding ${\sim}\, 10\%$ of these AGNs, being too shallow to detect the dim X-ray luminosity of the GW sources. Inspiralling binaries in an active phase with masses ${\lesssim}\,10^6\rm M_{\odot}$ at $z\,{\leq}\,0.3$ can be detected, as early as $10$ hours before the merger, by future X-ray observatories in less than a few minutes. For these systems, ${\lesssim}\,10$ AGNs are within the LISA sky-localization area. Finally, the LISA-Taiji network would guarantee the identification of an X-ray counterpart $10$ hours before merger for all binaries at $z\,{\lesssim}\,1$.
Comment: 25 pages, 14 Figures, To be submitted to MNRAS