I have noticed an issue with the helpful review filter which has been bothering me.

The helpfulness of the review is not based on the ratio of 'yes' to 'no' votes but rather purely on the total number of 'yes' votes. For example one review might have 72 out of 720 'yes' votes, making it 10% helpful, while another might have 71 out of a 100 'yes' votes making it 71% helpful. Yet the 72 out of 720 review wil still be rated as a more helpful review.
This skews the filter results by placing often useless, rubbish reviews that say uninformative things at the top which are not actually helpful to the decision to purchase.

Is it really so painful to just read a few more reviews or just go external?

Making "Helpful" reviews be based on a Yes to No ratio of votes is not helpful either. That heavily skews results, because fanboys of a given game will often mass downvote any critical review that is actually useful but is also honest about how bad a game is.

Yes, I do think GOG should base it on a weighted ratio (so as to not have the most helpful always be a 1/1 or 2/2) rather than mere upvotes. And biased voting happens in both directions, which review will end up on top currently is simply depends on whether the game has a larger fan or hate following. A relatively new review for an old game has virtually no chance of making it to the top, as most wont dig especially far; thus even if a greater percentage of readers upvotes a new contender than the reigning champ, not enough people will even see the contender for it to matter—the distance in terms of total upvotes will continue to grow. This can be even more of an issue if the old champ contains outdated information.

Oh yeah, and we should be able to see what we voted and change our vote, should we want to.

Yes Darvond it is really painful, so painful in fact that I went to the doctor who gave me one of those smiley face charts for pain rating. Not wanting to seem pretentious I selected the "severe pain" instead of the "extreme pain". Noticing the expression on my face caused by the entire 'helpful review' debacle and how it lined up with the chart, the doctor put me on a prescription of Darvocet, Vicodin, Percocet and Oxycontin. He also advised that I come here and annoy you with my observation on the review system. Yes that could happen. Perhaps the most helpful reviews should depend on a combination of weighting and the actual number of votes. The current system just isn't very helpful and often pins a review as 'most helpful' even though it essentially says nothing.

Statistically speaking there is a notion of a certain probability that users find a review helpful. That's the quantity you want to order them and you know the number of helpful/unhelpful votes. From that one can estimate the helpfulness probability as the ratio of the helpful votes to the total number of votes, exactly as the first poster here wants it to have.

To make it slightly better sort by highest estimated helpfulness-probability reduced by the estimated error in order to accommodate for the lower amount of knowledge about the true helpfulness of reviews with only a few votes. This should be a fairly standard problem that has already been solved and is used on many web pages.

As a formula: k is the number of helpful ratings on a review, n is the total number of ratings

estimated helpfulness is the ratio k / n

error in this estimated helpfulness is the square root of the inverse Fisher information which is something like: square root of (k/n(1-k/n)/n)

and if you computed estimated helpfulness minus error in the estimated helpfulness and simplify the equation a bit you get

(k - square-root-of(k/n(n-k)))/n

which is not difficult to compute.