Assessing correspondence between subjective reports and visual targets during attempted out-of-body experiences

Claims of veridical perception often accompany out-of-body experiences (OBEs), but systematic evaluation under controlled conditions remains difficult. We examined whether descriptions produced after attempted OBEs corresponded to concealed visual targets beyond chance using computational semantic similarity. Twenty-one individuals affiliated with an institute specializing in OBE training attempted to perceive a randomly selected target displayed in a separate, inaccessible room. Targets were drawn from a standardized database and remained unknown to participants and researchers. Ten participants reported an OBE, and 13 provided target descriptions for analysis. Participant descriptions and target captions were converted into high-dimensional vector embeddings using a pretrained language model. For each trial, cosine similarity between the participant's description and each of 100 candidate targets was computed; the correct target was evaluated against the within-trial decoy distribution using a rank-based metric (our primary analysis) and standardized z-scores (a secondary, descriptive analysis). Control analyses confirmed that the pipeline identified correct targets when semantic correspondence was present, though these controls do not establish sensitivity to naturalistic human language. No individual trial showed statistically significant above-chance correspondence, and a combined rank-based test across all 13 trials likewise showed no evidence of above-chance performance. Two participants spontaneously reported accurate details about experimenters' locations and activities in an adjacent room. Although the target-identification results did not support anomalous cognition, the spontaneous environmental reports may have methodological value and warrant systematic investigation.