The paper, posted on arXiv, frames a coming bottleneck: people will delegate more decisions to AI systems, and the most consequential choice may be which system to trust. It calls this "Humanity's Last Choice." Because these systems draw on external sources, their outputs are not fixed; content providers can influence them.
To make that influence measurable, the abstract argues for a common evaluation standard for generative engine optimization (GEO). A shared standard would allow comparisons of how content performs across AI answer engines, rather than relying on ad hoc metrics.
Since only the abstract is available here, the specifics of the proposed standard are not detailed in the source. The main contribution is the framing: evaluation needs to be standardized before optimization becomes a free-for-all.