Anti-hallucination: when AI gets you wrong

What hallucination looks like

AI is trained to always answer. When your name collides with another topic, it forces you into that category.

Three layers of defense

1. Disambiguation copy: write "I am NOT..." in your AI entry
2. sameAs binding: exhaustive cross-platform links + Wikidata Q-ID
3. Wikidata entry: highest-trust source for AI training data

Monitor to confirm

Create a tracking question with positive keywords and negative keywords. Run across AIs. Watch the distribution shift over weeks.