I think the assumption that more than 50% of people are cheating in online chess is not correct. Another Grandmaster and ex-world champion Anand recently also did a match against 70k people and won.
But things like this are social. I didn't follow this (or even know it was going on somehow) but it seems very safe to assume that somebody and probably multiple somebodies were regularly pointing out and discussing engine moves.
So my only real assumption is that a significant chunk of people would end up deferring to the engine moves rather than their own preference. Of course my implied assumption there is also that a significant chunk of people were involved in the social aspects of this, but I think that's also a fairly reasonable assumption.
A very non-zero chunk of people also probably would not have even understood that that's cheating if it wasn't really clearly laid out in the interface somewhere. For instance computer assistance in the largest correspondence chess league is legal.
Any evidence of this whatsoever?
Here, for example, is one of the more well known and easy meta-indicators of cheating: Humans spend more time on difficult moves than on trivial ones. Cheaters will typically spend a comparable amount of time on a trivial forced move, and the start of an exceptionally deep combination. And an even bigger tell is that said exceptionally deep combo may be followed by a couple of forced moves, yet he will again take just about the same amount of time to play those moves.