Instead, I rely on my training data to understand and generate text. This data includes massive amounts of text from books, articles, websites, and other sources.
When you ask me to spell a word, I use this data to find the most likely spelling based on the context of your question. I can also access and use external dictionaries and spell checkers to improve my accuracy.
So, while I don't spell "accurately" in the same way a human does, I do my best to produce accurate spellings based on my knowledge and the information available to me.