Context
An original research preprint introduces a citation-failure taxonomy, an iterative AgentGEO diagnosis-and-repair framework, and the MIMIQ document-centric benchmark with 204 webpages and 12,240 generated queries. In controlled simulated generative-engine evaluation, the authors report higher citation rates after targeted repair.
What changed
Rather than uniformly rewriting text, the study diagnoses failures in fetching/context, parsing and answer-generation stages, then selects targeted interventions and evaluates generalization on held-out queries. In one in-context simulation, reported citation rates were 79.52% for AgentGEO, 68.80% for AutoGEO, and 56.58% for unchanged pages.
Why it matters
Shows why AI citation measurement should distinguish source access, extraction and final citation decisions; content changes cannot solve every upstream or domain-level selection problem.
Visibility OS interpretation
Visibility OS interpretation: a useful research framework for stage-specific GEO audits and hypothesis testing, not evidence that the method increases citations in production ChatGPT Search, Claude Web Search or other commercial engines.
What remains uncertain
Can the improvements replicate on live commercial search products, organic retrieval pools and real user queries? How do failures vary by topic and existing citation saturation? The study reports negative or minimal effects for some health pages and calls for external validation.