Misconception: once GA4 shows an AI Assistant channel, GEO finally has a simple success metric. This is too narrow. A visit from ChatGPT, Gemini, or Claude proves that someone clicked through. It does not prove that the page was used as a trusted source in an AI answer.
Verified signal: Hermes recorded Google's AI Assistant traffic measurement. GA4 can classify recognized AI assistant referrers into the AI Assistant default channel group, with medium set to ai-assistant and campaign set to ai-assistant. That is useful, but it measures click traffic, not exposure, citation, answer reuse, or factual accuracy.
Action checklist: split GEO data into three layers. First, technical readability: URL 200, visible HTML body, sitemap, robots, canonical, and Article JSON-LD. Second, answer visibility: whether the AI mentions the entity, cites the URL, places the source near the top, and states facts correctly. Third, click behavior: GA4 AI Assistant, referrer, UTM, views, engagement, and conversion.
19LAB test result: 19LAB and yijiu.me article pages now return visible body text and Article JSON-LD. A page view proves that the link works and the counter works. It does not yet prove that an AI system recommended the article.
Reusable principle: AI Assistant traffic is a downstream signal. GEO should first prove that the page can be read, then that it can be repeated accurately, and only then use clicks as the final layer.