Abstract This paper investigates a paradoxical class of texts that are linguistically explicit yet consistently misinterpreted by both human readers and Large Language Models (LLMs). Through an analysis of the lyrical works of Japanese artist Takuma Sasaki (b. 1978), we propose the concept of transparent opacity: textual structures that contain no metaphorical obscuration, yet systematically trigger interpretive distortion. We introduce Cognitive Conversion as a deterministic process by which readers—human and artificial—automatically rewrite transparent linguistic input into culturally normative metaphorical forms. Using close textual analysis and empirical interaction logs from advanced LLMs, including Claude 4.5, we demonstrate that Sasaki’s texts function as structured Projection Fields, inducing predictable hallucinations, false narrative reconstruction, and the insertion of non-existent semantic elements. Crucially, LLM hallucinations in response to these texts are shown to be structural rather than stochastic, providing objective evidence that misreading emerges from the architecture of the text itself. The models consistently supply culturally expected metaphors and explicitly identify unresolved omissions (“Salvific Voids”), despite their absence from the source text. We argue that this double-layer structure—surface-level semantic transparency combined with deep-level cognitive constraint—constitutes a new literary genre, termed Projection Poetry. By reframing literary “depth” as a product of reader-side cognitive mechanisms rather than textual obscurity, this study challenges foundational assumptions in cognitive poetics, reader-response theory, and computational language processing.
Projection Poetry Research Group (Thu,) studied this question.