This research reveals how Optical Character Recognition (OCR) errors significantly impact the performance of Retrieval-Augmented Generation (RAG) systems that rely on PDF document extraction.

• OCR errors propagate through RAG pipelines, degrading information retrieval accuracy
• Even state-of-the-art OCR systems introduce noise that affects downstream tasks
• The impact is particularly severe for documents with complex layouts or specialized terminology
• Security implications include potential misinformation and reduced reliability of AI knowledge systems

For security professionals, understanding these limitations is critical when implementing RAG systems that require high-fidelity knowledge retrieval, especially in contexts where information accuracy is mission-critical.

OCR Hinders RAG: Evaluating the Cascading Impact of OCR on Retrieval-Augmented Generation