Content test
Island Test
Also: island test, semantic completeness test
AI extraction modules sever passages from their host document and insert them into a synthesized response. A passage that requires surrounding paragraphs to make sense loses its meaning when severed. The Island Test ensures the passage survives extraction with its facts intact.
The test is binary. The passage either passes the Island Test or it does not. There is no partial credit. Passages that fail the Island Test are systematically dropped by ChatGPT, Perplexity, and Gemini extraction models.
Cited facts
- Semantic completeness correlates r=0.87 with Google AI Mode citation frequency (Wellows 2026 ranking factor study).
- Passages must be strictly 134 to 167 words to optimize for the Island Test (Wellows 2026).