Workplace AI can rehearse cultural encounters—but only four papers directly studied the idea
A peer-reviewed review found a thin direct evidence base for using generative AI in intercultural training. Scenario practice is plausible; better cultural judgement is not yet demonstrated.
The short version
- What happened
- The review retained 48 topical publications from 2021 to July 2026, but only four—8.3%—directly addressed generative AI and intercultural learning.
- Who it affects
- Workers and students may receive culturally confident AI guidance that sounds specific while missing within-culture differences; human review and reflection remain necessary.
- What remains unproven
- Only four of 48 topical papers directly covered the GenAI–intercultural intersection
Keep in mind
Only four of 48 topical papers directly covered the GenAI–intercultural intersection
No causal estimate of training effectiveness
Original search and screening denominators were not preserved
No second human coder
Recent 2021–2026 window excludes foundational earlier research
The publisher-supplied placeholder DOI was unresolved at verification time
Source analysis · 2 linked sources. Primary analysed source date: 26 October 2026.
Read the full report · 3 min The evidence, methods, implications and limitations
The direct evidence is four papers, not a validated training method
A peer-reviewed review from Mahasarakham University in Thailand found that the case for generative AI in intercultural training is much thinner than the enthusiasm around it. The authors froze a bibliography of 58 verified sources: 48 topical publications and 10 methods or indexing references. Of the 48 topical papers, only four directly examined the intersection of generative AI and intercultural learning. The rest came from adjacent work on cultural intelligence, metacognition, AI-assisted learning or cultural bias.
That distinction matters for people learning to work across languages and cultures. A model can generate role-play, alter power relationships or provide quick feedback, but fluent output may also flatten culture into stereotypes. The review therefore treats AI as material to question and revise—not as a cultural referee. Trainers should keep human facilitation, ask learners to expose assumptions, use culturally diverse review and measure unaided transfer after the tool is removed.[1][2]
Useful design advice, with an incomplete review trail
The evidence check is limiting. The review covers publications from January 2021 to 27 July 2026, but did not preserve the original counts for retrieval, deduplication, title screening or full-text screening. It is therefore a focused bibliometric review, not a fully reproducible systematic review. One researcher performed the primary-stream coding; deterministic sensitivity analysis produced 97.9% agreement, but there was no second human coder.
The authors reported no funding or competing interests and disclosed AI-assisted language editing. Confidence would rise through preregistered comparisons of AI-only, human-facilitated AI and conventional training, followed across sites, languages and real workplace interactions. Until then, the practical message is restrained: use generated scenarios to practise checking judgement, but do not claim the tool has taught cultural competence.[1][2]
The distinction that matters
What the review supports—and what it does not
The paper moves from a broad adjacent literature to a narrow claim about direct evidence.
Recent literature
Classify 48 topical publications
Map cultural intelligence, metacognition, AI learning and bias
Direct intersection
Identify four relevant papers
Find plausible uses, not proven training effects
Next test
Compare AI-only, facilitated and conventional training
Measure unaided behaviour and transfer over time