Background <p>Collaborative clinical reasoning (CCR) is essential to safe and effective practice in intensive care units, where decisions are shaped by uncertainty, high cognitive load, and interprofessional dynamics. Although CCR is widely recognised as a collective cognitive process, little is known about how clinicians learn to reason together or how reflective educational interventions may influence the language and social organisation through which CCR is enacted.</p> Methods <p>This exploratory mixed-methods Groundwork study examined the association between a video-facilitated reflective (VFR) inquiry, informed by Video Reflexive Ethnography (VRE) principles, and the linguistic features of CCR within a neuro-intensive care unit (NICU) in Taiwan. Twelve clinicians (4 attending physicians, 2 resident physicians, 5 registered nurses, and 1 respiratory therapist) participated. Eight bedside ward-round sessions were video-recorded across two phases (4 pre-intervention; 4 post-intervention). Audio was transcribed in Mandarin and translated into English by bilingual clinicians. English transcripts were analysed at the session level using Linguistic Inquiry and Word Count (LIWC-22). Five a priori focal constructs—analytical thinking, clout, authenticity, tentativeness, and insight—were selected on theoretical grounds linked to dual-process reasoning, epistemic stance, and team cognition; 5 additional categories were retained as descriptive context. Independent-samples Welch t-tests, Hedges’ <i>g</i> effect sizes with 95% confidence intervals, and Mann–Whitney <i>U</i> robustness checks were used to compare pre- and post-intervention sessions. Quantitative trends were interpreted alongside qualitative reflections from retrospective think-aloud sessions in which participants reviewed selected video excerpts of their own practice.</p> Results <p>Of the 5 pre-specified focal constructs, only insight showed a robust post-intervention elevation, consistent across parametric (Welch <i>t</i>(5.87) = -3.78, <i>p</i> = .010) and non-parametric (Mann–Whitney <i>U</i> = 0.0, <i>p</i> = .029) tests, with Hedges’ <i>g</i> = 2.33 (approximate 95% CI [0.53, 4.12]); this finding survives a Bonferroni-adjusted threshold of <i>α</i> = .01. Tentative language showed a parametric trend in the predicted direction (<i>t</i>(5.08) = -2.57, <i>p</i> = .049; <i>g</i> = 1.58, 95% CI [-0.01, 3.17]) that did not survive non-parametric testing (Mann–Whitney <i>p</i> = .114) or multiplicity considerations, and is therefore reported as suggestive rather than detected. No effect was detected for clout (<i>g</i> = -0.18), authenticity (<i>g</i> = 0.41), or analytical thinking (<i>g</i> = 0.47); all confidence intervals crossed zero. Confidence intervals are wide, reflecting only 4 session-level observations per phase. During reflective sessions, participants described heightened awareness of their own communicative habits, greater openness to uncertainty, and increased willingness to invite alternative perspectives from other team members.</p> Conclusions <p>In this exploratory single-site study with only 4 session-level observations per phase, video-facilitated reflective inquiry was associated with elevated use of insight-related vocabulary in subsequent ward-round talk, while a parametric increase in tentative language did not survive non-parametric robustness testing. No effect was detected for clout, authenticity, or analytical thinking. Wide confidence intervals, translation-related validity considerations, and the absence of a controlled comparison group warrant caution; the findings are hypothesis-generating rather than confirmatory. The combined use of computational linguistic analysis and reflective dialogue offers a tractable approach to surfacing subtle changes in collaborative reasoning that should be tested in larger, multi-site samples.</p>