Correlation vs causation is the difference between two things moving together and one of them causing the other. In AI visibility it is the most common reasoning error: a factor is found more often on cited pages, so it is assumed to cause the citation, when it may simply travel alongside the real cause.
The worked example is structured data. Pages cited by AI are far likelier to carry JSON-LD schema, a strong correlation, yet a controlled 2026 experiment that added schema to matched pages found no citation uplift. The resolution is that well-built sites tend both to add schema and to earn citations, so schema correlates with citation without causing it. The same trap lurks behind claims that a word count, a freshness date, or any single on-page trait "gets you cited."
Telling them apart takes a controlled comparison: change one variable, hold a similar control group unchanged, and see whether the treated group moves beyond the control. A raw "cited pages have X" count never can, because it cannot separate X from everything else those pages share. Reading GEO research through this lens, favouring controlled evidence over correlational counts, and hedging single-vendor findings, is what keeps a content strategy anchored to signals that actually move citations rather than ones that merely accompany them.