Comment by vianneychevalie

A chain of LLMs can work in that regard, using intermediary prompts that feed answers to the next prompt. Make the LLM build a list of sections, then make it fill them with examples, then make it enrich the text. Maybe a last layer for error correction, clarity, removing mentions of "as an AI model", etc.