Answer engines typically name a handful of sources per reply. GEO targets that citation slot with answer-first copy, concrete statistics, named sources, question-shaped headings, and machine-readable files such as llms.txt. The same techniques are used by marketers and, in 2026 reporting on the Hanover Institute, by state-funded content operations. GEO is a retrieval-time practice; it is not the same as poisoning a model's training weights.