Introduction: “Continuum” approaches to psychosis have generated reports of similarities and differences in voice-hearing in clinical and non-clinical populations at the cohort level, but not typically examined overlap or degrees of difference between groups. Methods: We used a computer-aided linguistic approach to explore reports of voice-hearing by a clinical group (Early Intervention in Psychosis service-users; N = 40) and a non-clinical group (spiritualists; N = 27). We identify semantic categories of terms statistically overused by one group compared with the other, and by each group compared to a control sample of non-voice-hearing interview data (log likelihood (LL) value 6.63+=p < .01; effect size measure: log ratio 1.0+). We consider whether individual values support a continuum model. Results: Notwithstanding significant cohort-level differences, there was considerable continuity in language use. Reports of negative affect were prominent in both groups (p < .01, log ratio: 1.12+). Challenges of cognitive control were also evident in both cohorts, with references to “disengagement” accentuated in service-users (p < .01, log ratio: 1.14+). Conclusion: A corpus linguistic approach to voice-hearing provides new evidence of differences between clinical and non-clinical groups. Variability at the individual level provides substantial evidence of continuity with implications for cognitive mechanisms underlying voice-hearing.