Misconception: one sharp keyword or headline is enough for GEO. That is still an SEO habit. AI search does not only match one phrase. It can expand a user question into definitions, risks, steps, comparisons, evidence, alternatives, and verification checks.
Verified signal: Google's AI feature documentation mentions query fan-out, and Hermes turned that into a sampling method. The right monitoring unit is not one keyword. It is main intent, fan-out subquestions, source pool, and answer facts.
Action checklist: before publishing, write 6 to 10 fan-out questions for the article. Make sure the body answers at least three of them in extractable paragraphs. Track which subquestions are definition, checklist, risk, comparison, verification, source, or next action. Then sample AI platforms with those subquestions instead of only searching the exact title.
19LAB test result: each 19LAB / yijiu.me GEO article should become a small answer asset. If the article only argues one point, it cannot cover the source pool that an AI system may assemble.
Reusable principle: GEO content should be designed as a compact answer cluster. The headline attracts attention; the fan-out structure makes the article reusable by agents.