Post-hoc attribution is the process of identifying, retrieving, and attaching verifiable supporting sources or citations to specific claims in a computer-generated text after that text has already been produced. Unlike retrieval-augmented systems that consult reference documents before or during text generation, post-hoc attribution evaluates completed model outputs by decomposing them into distinct factual statements and mapping each claim back to supporting external evidence. This technique is used to verify the factual accuracy of generated content, mitigate unsupported statements or hallucinations, and provide an auditable trail of evidence that enhances transparency and trust in artificial intelligence systems.
