Edit: The duplicate Q&A on StackOverflow has answers to this question (as pointed out by Tianxiang Xiong in the comments).
My answer below is no use, as you evidentially can't get the desired result that way. I'm only not deleting it because the information it does give is somewhat relevant, and might still prove interesting information to someone.
Personally I do just edit the merge buffer directly, but you can teach Emacs what to do if you have a standard requirement.
Combine Nth diff regions of buffers A and B and place the combination in C.
N is a prefix argument. If nil, combine the current difference regions.
Combining is done according to the specifications in variable
Pattern to be used for combining difference regions in buffers A and B.
The value must be a list of the form
(STRING1 bufspec1 STRING2 bufspec2 STRING3 bufspec3 STRING4)
where bufspec is the symbol A, B, or Ancestor. For instance, if the value is
'(STRING1 A STRING2 Ancestor STRING3 B STRING4) then the
combined text will look like this:
diff region from variant A
diff region from the ancestor
diff region from variant B
Note that when you think it's doing nothing, it's actually combining the variants using the normal conflict markers. i.e. If you had already selected A or B, then by default + is a way to restore the original conflict.