That is true of traditional chess engines, but no longer true for neural net (self-taught) engines that learn the rules of chess by simulating millions of games against itself.

I've played an early version of one of those (its playing strength is roughly proportional to the amount of training it has done) and it was quite human-like in its behavior, even dropping pieces and making very human-like mistakes.

Look up AlphaZero and the open source port LeelaChess :) Once those become commercially available, it's going to be very hard to detect engine usage.