BACKGROUND: Selecting optimal tasks for language mapping in neurosurgical patients poses challenges that are exacerbated by mismatches in practice between presurgical and intraoperative evaluations. To help align practices, we evaluated a functional MRI version of a semantic association task increasingly used during intra-operative assessment of awake neurosurgery patients. Using a recently proposed consistency mapping approach, we characterise task fMRI activation reliability across individuals, visits, and scan cohorts. METHODS: FMRI data were acquired during an adapted Pyramids and Palm Trees Task (PPTT) in 15 healthy controls and 54 pre-surgical patients with a glioma. A new implementation of threshold-weighted overlap mapping (TWOM) was used to evaluate: 1. inter-individual variability in task activations among individuals; 2. test-retest variability in controls scanned twice (16 ± weeks apart); 3. between-scanner reliability across two patient cohorts scanned on a 3 T Siemens Prisma (n = 27) or Verio (n = 24) scanner using standard (TR = 3 s, voxel size 3 × 3 × 3 mm) or advanced (TR = 0.93 s, voxel size 2x2x2 mm) fMRI acquisitions, respectively. RESULTS: Task-related activations in the core language network were highly consistent between individuals and across test-retest sessions. Several brain regions showed variable activations, reflecting atypical language dominance (confirmed during neurosurgery), or differences in regional involvement during semantic processing. CONCLUSION: The PPTT engaged widespread brain networks including but not limited to regions implicated in semantic processing. Overlap mapping is a powerful way to visualise meaningful variations in neural processing at the individual level, supporting alignment of pre- and intra-operative mapping for any given task.