Based on its sequence similarity to Apobec1, which deaminates Cs to Us in RNA, activation-induced deaminase (AID) was originally proposed to work on RNA. But definitive experiments in E. coli demonstrated that AID deaminates Cs to Us in DNA. The authors of the paper expressed AID in bacteria and followed mutations in a selectable gene. They found that AID expression increased mutations about fivefold above the background level in the absence of AID expression. More importantly, they found that 80% of the induced mutations were G→A or C→T. Does this fit with your expectation if AID-induced mutations arose by deamination of C to U in the DNA? Imagine what would happen if the G:U mismatch created by AID was replicated several times; how would the sequences of the final mutations relate to the original G-C base pair?