I'm not sure if you saw the portion of our article that develops the estimate and presents the methodology for the estimate. If not, that might be of interest.
As you say, it's hard to make a precise estimate. There are important pieces of data uniquely within Google's custody, and Google isn't talking. But in these circumstances, I do feel it's appropriate to make a good-faith estimate. If you think our numbers are in error, feel free to identify which specific numbers you think are off, in which direction, and for what reason. But realize that for every number you think is too high, there is likely to be another that might be too low. (We discuss some of these complications in the page linked above.) I don't think it's clear from first principles that our estimate is biased in one way or the other.