Free-to-play puzzles collect enormous numbers of ratings, most of them from prompts shown immediately after a player finishes a level and feels good. That is not fraud; it is a well-timed question. It does mean the headline average measures satisfaction at the happiest moment of the experience rather than the whole of it.
1. Sort by recent, then read the low ratings
Old complaints are noise — they may describe a version that no longer exists. A cluster of one and two-star reviews from the last fortnight, all mentioning the same thing, is a real signal: an update that broke saves, an event that failed, a purchase that never arrived.
2. Look for developer replies
Not their content, which is usually a template, but their existence and their date. A studio replying to reviews from last week is a studio still staffing support. A game whose last reply is from two years ago is coasting, and coasting free-to-play games break quietly after operating system updates.
3. Ignore reviews about difficulty, read reviews about money
"Level 340 is impossible" tells you the difficulty curve is doing its job. "I bought the bundle and did not receive it" tells you something about the company. The second kind is rarer and much more informative.
4. Check whether the reviews describe the same game you are looking at
Publishers sometimes reuse a listing: a game is rebuilt, rebranded or replaced, and the old ratings carry over. If the reviews from a year ago describe different mechanics or a different art style, the average belongs to a game that no longer exists.
What about the summary the store writes?
Both stores now generate short summaries of review themes. They are genuinely useful as a shortcut and genuinely unreliable at nuance — they will happily report that "players enjoy the graphics" while a hundred recent reviews describe a crash on launch. Use them to decide what to read, not as the reading.
The most useful review is your own, ten minutes in
Nothing here replaces installing the thing. Both stores let you remove an app minutes later at no cost. Ten minutes will tell you whether a board is readable, whether the animation is too slow, and whether the first thing the game does is ask for money — all things that no star average has ever communicated.