High-quality documentation is essential during the initiation phase of IT projects, yet producing standards-based artefacts remains time-consuming and error-prone. Recent advances in large language models (LLMs) offer opportunities to automate document authoring, but unstructured prompting often leads to structural inconsistencies, hallucinations, and weak alignment with business semantics. This paper introduces a meta-prompting approach so called “GenAI for GenAI” to guide generative AI in creating project documentation. The proposed “generation fountain” workflow uses one LLM to design precise prompts for another, while a semantic retrieval layer injects relevant business data into each instruction. A hybrid architecture supports local execution, minimizes context size, and applies strict rules to prevent unsupported content. The solution has been implemented in a prototype that generates IT project management plan baseline components and can also be applied to other types of documentation. By combining meta-prompt engineering, business data semantics, and controlled text generation, the approach demonstrates how generative AI can reliably support standards-based IT project documentation, even in resource-constrained environments.