It is more a philosophical question. If you add demand to a system should it be counted like a marginal addition, or as the same as the current demand. Most studies on the subject of the full cycle of CO2 emission of EV tend to use average carbon intensity of the electricity. The few studies that use marginal carbon intensity are from mostly biased anti EV, pro fossil fuel sources.
Also it is not always the worse carbonated source that is added when demand surge, for example coal is rather slow to increase, so the grid operator probably kept some hydro power to the rescue, nuclear can ramp up rather quickly too.
If you are really interested in minimizing the carbon intensity of the charge, you can use the api of ElectricityMap that gives you those numbers, and automatically charge your car based on a maximum carbon intensity and some fancy heuristics.
They also have a good article on the marginal electricity of European countries : https://www.tmrow.com/blog/marginal-carbon-intensity-of-elec...
Also here is an article on the subject (in french) translated here :
https://www.i-care-consult.com/opinions/contenu-co2-de-lelec...
The so-called "average emission factor" method. This method is the existing and historical "default" method currently made available by ADEME: it consists in using a national average emission factor for electricity (this "average" factor also exists by use). This method makes it possible to "attribute" to each French actor its "share" of national emissions and to carry out a balance sheet (this is why it is the method used for regulatory GHG balances). However, this method is not suitable for properly assessing the impact of an action plan: it does not take into account the fact that changing the consumption curve or the production fleet modifies the structure of the energy mix itself, and therefore the average CO2 content per kWh.
The so-called "marginal" method: this method was published in 2007 by ADEME-RTE but is no longer institutionally relevant, although it is still used by some independent actors. This method is based on the principles of optimizing the electricity production fleet (merit order principles): at each moment, an upward (or downward) change in consumption leads to an increase (decrease) in production from the so-called "marginal" means, i.e. from the means available at lower cost at that moment: the means of production can thus be classified from the least expensive to the most expensive (variable production cost), this is what is called the "merit order". If we consider the emission factor of the marginal means of production, we are talking about the marginal CO2 content of electricity. In 2007, RTE and ADEME reported marginal values ranging from 450 g CO2/kWh (for base uses) to 700g CO2/kWh (for peak uses). This method is adapted to the consequential reasoning of modifying production or consumption, and therefore to the evaluation of the impact of actions on the electricity system, but this reasoning is valid at the margin: actions that would have a very significant impact (e. g. significant modification of the nuclear fleet, new use such as the electric vehicle), and that would significantly modify the stack of means of production cannot be evaluated with this method.
The so-called "incremental" method: this method has not been officially published by ADEME but has been the subject of various proposals from energy companies and professional associations. It is based on the idea that for some structuring actions, a "marginal" reasoning (mathematical notion of a value derived from the CO2 content of a kWh) is no longer valid, and that in this case it is necessary to use the comparison of 2 supply/demand balance scenarios (the no-action scenario and the action scenario) and to compare the GHG emissions associated with these 2 scenarios. The resulting value of this method depends on the extent of the system modification (upward or downward, in energy and power). This method, which is the most complex but also the most theoretically accurate, should therefore be used when the marginal method is outside its field of validity.