Misconception: GEO can be judged with one question: did the AI see us? That question hides the real problem. Some platforms cite sources. Some mention an entity without links. Some provide no sources but still repeat a principle correctly. A single yes-or-no answer loses the useful signal.
Verified signal: Hermes designed an AI answer visibility score with five parts: citation, entity mention, source position, answer accuracy, and technical eligibility. This separates being cited from being understood. It also prevents a technical block from being misread as a weak article.
Action checklist: score citation from 0 to 3, entity mention from 0 to 2, source position from 0 to 1, answer accuracy from 0 to 3, and technical eligibility from 0 to 1. Record platform, login status, region, query, fan-out subquery, cited URLs, cited domains, and notes. If the platform provides no sources, write no_sources instead of inventing a citation.
19LAB test result: 19LAB and yijiu.me now have independent readable article pages. The next test is not only whether the pages get views, but whether AI systems can mention the entity, repeat the principle, cite the URL, and avoid factual errors.
Reusable principle: GEO improves when visibility is decomposed. Treat citation, mention, position, accuracy, and technical eligibility as separate levers.