Puzzles are one of the most recommended ways to improve at chess, and also one of the most misused. Grinding a random puzzle feed for an hour, rated 800 one moment and 2200 the next, feels productive because you are doing something chess-shaped. Whether it actually makes you better at your own games is a separate question.
Random puzzles train pattern recognition in general
A large, varied puzzle set is genuinely good for one specific thing: broadening your library of tactical patterns so you recognize more shapes on the board. That has real value, and it is part of why Rookie's own puzzle trainer pulls from Lichess's public database of real game positions rather than inventing artificial ones.
What it does not do well is target the handful of mistakes that are actually costing you games. A random feed treats a fork you would never fall for the same as a back-rank pattern you have missed four times this month. Both count as one puzzle solved, but only one of them is addressing a real leak in your own play.
The puzzles that transfer best are your own
A puzzle built from a position you actually reached, in a game you actually played, against the exact mistake you actually made, transfers to your future games far more directly than a puzzle from a stranger's game a decade ago. You already have the context. You already felt the moment go wrong. Solving it again, and then solving it again a week later once it resurfaces, is what turns a one-time correction into a habit.
The honest answer, then, is to do both. Use a broad puzzle feed to keep building general pattern recognition, and use puzzles built from your own recurring mistakes to close the specific gaps that are actually holding your rating back.
Rookie gives you both in one place: a free puzzle trainer pulling real tactics from Lichess's public database, and, once you sign up, puzzles built directly from the mistakes in your own games, resurfaced on a spaced-repetition schedule until they stop happening.