Past and future performance do correlate in many cases, but "Google and Microsoft do it" has never been a valid argument for the correctness of anything. I hear that in Imperial China one had to pass a classical literature exam to get civil service jobs. There is a great deal of arbitrary convention in the things done by big companies, this is not evidence of optimality.

When people are at a loss for how to make a hiring decision and they have a giant pile of applications in front of them, they are liable to just choose whatever filter is available. It might not be any better than reading tea leaves, but at least the decision gets made. This is rational behavior when the cost of devising or applying a better criterion is high, but it doesn't mean the resulting criterion is the best thing possible.