Known phrases are swapped outright, and longer words pick up the -diddly- infix at random, which is what keeps the effect from feeling mechanical.
The character's speech habit is a genuine phenomenon called expletive infixation, inserting a chunk into the middle of a word rather than at either end. English is unusual in allowing it at all, and it follows a strict rule: the insertion goes before the stressed syllable, which is why fan-diddly-tastic works and fantas-diddly-tic sounds wrong. The nonsense syllable is what removes the aggression from a pattern normally used for swearing, which is the joke, an infix built for profanity, filled with something harmless.
This tool inserts the infix in roughly a third of long words rather than all of them, and near the middle rather than at a computed stress point. Applying it to every word makes the output unreadable within a sentence, and finding the stressed syllable properly needs a pronunciation dictionary. Random placement gives the effect its unpredictability, which is closer to how the character actually speaks. The habit surfaces occasionally rather than mechanically. Known phrases like hello and okay are swapped outright, since those have established forms.
Because infixing every word makes a sentence unreadable. Roughly a third of longer words are affected, which is closer to how the habit actually surfaces in speech.
It is approximate. Correct infixation goes before the stressed syllable, which needs a pronunciation dictionary. This inserts near the middle instead.
The placement is deliberately random, so running the same text twice gives different output. That unpredictability is part of the effect.