Academic studies are hampered by the fact that a lot of them only use, for example, the small sample of students in Lyon for France, one small sample of half Catalans/half Andalusians for Spain, and a small sample from Thessaloniki for Greece. At least for Italy they usually have three samples.
However, there are problems with the ancestry sites as well. 23andme uses its own testees as part of the reference sample, which is helpful in some ways, but, for example, they're undoubtedly flooded with Americans with ancestry from southwestern Germany, which makes the French and Germans look almost identical, and they also have a lot of Southern Italian/American customers, so the "centrum" for Italy drifts south.
Deciding which areas should be in which cluster is more subjective than you might think. The Balkans are a perfect example. Where do you put Greece? Ancestry used to have a Southern Italian/Greek cluster. It made some sense. Now, looking at their results, you wouldn't know how similar some of the Balkan countries also are to Italians.
The "truth" is rather amorphous, and can only be gleaned from a gestalt of the meaning of all these different tests. You have to be able to think "globally" not literally, and the conclusions are difficult to put into a few sentences.