In that case, with what I know about the corpus used to train ChatGPT, I object to the claim these are better synonyms. They are certainly the statistically most likely to be used. But if I wanted to choose a synonym based on frequency of usage, I'd grab some data about word frequencies[1], throw it in a thesaurus tool and sort by frequency.
That's true, but you can prompt the LLM to sort by other criteria and it will happily oblige. e.g. number of syllables, how posh the word sounds, or in this case relevance to the context.
1 https://www.wordfrequency.info/