Intelligent Scientific Literature Analysis with Generative AI: A Literature Review and Conceptual Framework

New research papers are published every day, and students and researchers cannot search, read, compare and summarize the relevant ones by hand without losing a lot of time. Generative AI tools, mainly large language models (LLMs), retrieval-augmented generation (RAG), natural language processing (NLP) and AI agents, could cut some of that effort by finding papers, writing summaries, answering specific research questions and pulling out key findings. This paper looks at five works that run from Kitchenham's systematic review method to the LitLLM toolkit and compares their methods, findings and limitations. Six gaps came out of that comparison. Models sometimes invent things and cannot compare papers well. Large collections slow them down. Specialised fields are another weak spot, and they never show how they reached an answer. In the end a person has to check the result anyway. The paper then proposes an eight-stage conceptual framework that goes from database collection through NLP, retrieval, LLM reasoning and generative summarization to a final analysis report, along with evaluation measures and ethical points. The framework is only a concept; it has not been built or benchmarked.