Using fuzzy matching to determine whether segment(s) of responsive content, that is generated using generative model(s), match segment(s) of additional data

Topics: AI Mode, AIOverviews, LLM Readability, LLMO / GEO, Passage based retrieval, Retrieval Augmented Generation (RAG)

This patent by Google LLC describes a system that uses fuzzy matching (based on edit distance calculations) to check whether content generated by a generative AI model closely matches segments of other data sources — such as the user’s own documents, search result documents obtained via retrieval augmented generation (RAG), or the model’s original training data. Depending on whether a match is found and what type of source it matches, the system dynamically decides whether to modify, attribute, truncate, omit, or regenerate segments of the AI-generated content before presenting it to the user. The goal is to ensure proper attribution of AI-generated content while conserving computational resources by prioritizing which data sources to check first and skipping unnecessary processing steps.