The "GPT" in ChatGPT stands for Generative Pretrained Transformer, and the "T" is this architecture. A transformer turns your text into tokens, represents them as vectors, and uses attention to decide which earlier tokens matter most when predicting the next one. Repeat that prediction loop and you get fluent text.
The useful takeaway for content is what attention rewards: clear referents, consistent terminology, and answer-first structure are easier for the model to resolve than ambiguous, scattered prose. You are not optimizing the architecture, you are making meaning easy to compute, which is the same thing good writing has always done.