Context
Counter-GEO-Bench introduces a controlled benchmark pairing 247 human-verified, quality-gated queries with information-preserving and information-distorting GEO rewrites, evaluating defenses with attack success rate, false-positive rate and answer quality across three victim LLMs.
What changed
The research adds a dedicated benchmark for evaluating defenses against GEO-based misinformation. In the reported experiments, three off-the-shelf defenses reduce attack success only modestly, while the proposed C-GEO Guard baseline reports a 47.6% relative reduction with near-zero utility loss.
Why it matters
The benchmark makes adversarial source selection measurable and distinguishes conventional safety filtering from defenses designed specifically for information-distorting GEO.
Visibility OS interpretation
Visibility OS interprets this as evidence that AI visibility measurement should distinguish benign optimization from manipulative visibility and include defensive retrieval/source-selection behavior as part of the measured discovery system.
What remains uncertain
How well do these defenses generalize to production retrieval stacks, unseen GEO methods, changing model families and multilingual queries? What defensive interventions preserve legitimate publisher visibility while suppressing manipulative rewrites?