PIE reconstructions are very interesting peaces of linguistic, but they seems often mistaken. One great analogy, I first saw presented in some Linguisticae[1] video I think, is "what if we had no direct trace of Latin and we were looking to recreate proto-Romance roots." Of course Latin itself refers to very wide set of linguistic practices, with all the diversity we can imagine through time, space, individuals and even for a given individual there are difference as they age and depending of context they will use different sociolects and language register, plus of course not everyone is mono-linguistic.
That's what I wonder, whether there has been any blind backtesting of the methodology itself to see how reliable it even is. Reconstructed proto-languages tend to be overly complex and unnatural.
There have been attempts to recreate (vulgar) Latin from modern day Romance languages, as well as using older forms of these languages to reconstrct what's known as Proto-Romance.
My recollection is that the complexity went the other way; Latin was more complex than the reconstructed languages, especially if the reconstruction didn't include Romanian, because the modern Romance languages became simpler over time in similar ways.
It's clear that the result is useful for understanding features of the ancestral language, but it's not perfect, and never will be.
On the other hand, comparative linguistics came long before genetics, and it is this field that first noticed a connection between the Indo-European languages.
Archaeological and especially genetic evidence now show the peoples of this language family (mostly) have shared (though distant and diluted) ancestry, so the field was broadly correct in noticing a connection.
[1] https://www.youtube.com/@Linguisticae