In this study, we investigate the impact of scattering-related XCO2 uncertainties of CO2M on retrieving CO2 emissions from cities. XCO2 uncertainties are provided by CHE deliverable D2.5 for a retrieval without multi-angle polarimeter (MAP) and are based on a neural network. We perturb synthetic CO2M observations according to these uncertainty estimates and assimilate them in a CO2 flux inversion system to update prior flux estimates. We analyse single overpasses of Berlin, Beijing and Shanghai in winter and summer in the year 2015. Scaling whole city emissions yields uncertainties of 11–25% in Berlin and -1–9% in the Chinese cities, suggesting promising performance for snapshots of emissions derived from single overpasses. We also try to constrain individual emission hotspots in Berlin based on the high-resolution flux dataset from D2.3. With flux errors of several tens of percent, this is less successful than constraining city totals. Reasons for the larger uncertainties of individual hotspots are conflation of several sources and sensitivity to spatial variations of XCO2 biases on the scale of individual CO2 plumes. The results for Berlin are similar to those from the H2020 project AeroCarb. However, in contrast to our study, AeroCarb found very large flux uncertainties in Beijing for the XCO2 retrieval without MAP. The main reason for this discrepancy is that CO2M uncertainties appear to have been overestimated in AeroCarb, as an update brings them closer in magnitude to the results of the neural network D2.5 uses. However, the two methods do not agree on the spatiotemporal distribution of systematic XCO2 errors, and XCO2 error variability was estimated higher outside of the small city domains we assimilated. Assimilating this higher error variability would likely yield larger flux errors than in our cases. Therefore, the differences between the XCO2 uncertainty estimation methods should be reconciled. While modelled aerosol optical depths, which are used to estimate CO2M uncertainties here, are not representative for heavily-polluted episodes in China, they are close to average conditions in the three cities and thus representative for a high number of days throughout the year 2015. Our modelled CO2 plumes from Beijing and Shanghai are on the high end of the few signals from these cities that are to date observed by OCO-2 and OCO-3. Therefore, these large signals may be rare, and inferring annual total emissions may be difficult. Overall, the ability of CO2M to constrain CO2 emissions from large cities in the presence of scattering uncertainties seems promising in favourable conditions. However, differences between XCO2 uncertainty estimation methods should be reconciled to boost confidence in flux uncertainty estimates. Constraining smaller sources, fluxes on regional scale and annual emissions likely requires minimizing the XCO2 uncertainties we investigated further, as well as addressing other error sources in CO2 flux estimation.