I know everyone is unique and has their own opinions on things, but let us be honest. Regression to the Mean is a thing. And as Francis Galton showed (the father of Eugenics, so not exactly someone I would care to know personally) in his research in 1906, an individual can be wrong, but a large group will be much closer to the correct answer than any individual.
In the case though, the sample size and population matter. If a game has over 100 reviews and there is no evidence of monkey business, I think I will trust the review. That doesn’t mean I will buy a racing game because I don’t like racing games, but if it scores an 89% and had a large enough sample (say, over 40) then I would argue that it is a pretty good race game.
Now, the goal would be to prune out as many cheaters as possible. Limiting the pool to people who purchased the game is a good start but I would also love to see the game reviews limited to people who have played a few hours as well (although that might skew positive since bad games will have fewer people with negative reviews because of the lack of hours).
And keep in mind, this is a self-selecting sample, so it’s never going to be great. Now, if Valve started sampling people at random instead of people volunteering, now that might be cool.