Common letters are not the same as useful letters. The graph answers a different question.
Every word game attracts the same advice: play the most common letters first. In a word ladder that advice is almost useless, because the question is not which letters are common but which letters connect.
A common letter can sit in a dense family that leads nowhere. A rarer letter can be the single bridge between two families, which makes it the most valuable letter on the board. Frequency tables describe how often a letter appears in text, not how often a letter change produces a legal new word. The graph is the thing that matters, and the graph is local.
This has a practical consequence for move order. The first move should usually change the letter that produces the most useful neighbours, and that is a property of the specific word on the board rather than of English overall. Two starting words with identical letter frequencies can have completely different neighbourhood sizes.
The way to build intuition here is to count neighbours rather than to count letters. Pick a word, list its legal one-letter neighbours, and look at how they are distributed. Some letters produce four or five, others produce none. After a couple of weeks the distribution starts to feel familiar, and the frequency table loses its charm.
The idea is worth testing on a real puzzle, because it changes what a player looks at in the first ten seconds. The unlimited mode is the cheapest laboratory since rounds can be replayed, and practice as long as you like lets the same word be examined from every letter change until the neighbourhood is clear.
The replacement habit is simple: keep a note of how many neighbours each starting word actually has. Two weeks of that produces a better guide to English than any frequency table, because the numbers come from the graph the game is played on rather than from the language in general.