Accepting for a moment that the data is correct, the problem they have is that considering the different emotional states, and other dynamics, between the two samples, a 48% match is not indicative of dissimilarity, and I’m sure they know that.
Notice that they tell you the needed value for similarity under ideal conditions, but they never tell you the needed values for dissimilarity.
This is pure spin, and if I had an appropriate degree and the medium to dliver it. I would publicly humiliate these hacks in no uncertain terms.
Again, they lie to further an agenda.