Language differences impact LLM negotiation outcomes
In short: Research shows that identical persona instructions produce different behavioral outcomes in LLMs when switching from English to Korean. While buyers followed instructions similarly, sellers became significantly more accommodating and polite in Korean, leading to higher agreement rates and lower prices.
This summary was generated automatically by AI from arXiv cs.CL's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1Korean negotiations reached agreement 24.8% more often than English in Gemini 3.6 Flash tests.
- 2Sellers used significantly more politeness mitigation in Korean (11.7% vs 3.6% in English).
- 3Results replicated directionally with Qwen3.7 Flash.
Why it matters
This implies that a prompt-defined social policy is not globally consistent. For businesses using multilingual AI agents, the same system instructions might result in different levels of service or negotiation aggression, requiring language-specific tuning.
What it means for AI agents and contact centers
If you build multilingual voice agents, assume that 'headstrong' or 'cooperative' personas may soften or harden depending on the target language's linguistic norms. When benchmarking your agents for customer service or sales, perform language-specific evaluation rather than trusting that prompts translate perfectly into behaviors.
🧪 Worth Testing
Testing is essential to ensure that your agent's negotiation or sales style remains compliant with company policy across all supported languages.
Sources
- arXiv cs.CLOfficialPrimary sourceOriginal article →„Cross-Lingual Persona Effects in LLM Negotiation“3 Nov 2026, 07:00Licence: arXiv metadata CC0; papers under their own licences · our summary (content changed)
- Published by source
- 3 Nov 2026, 07:00
- Found by our system
- 3 Nov 2026, 11:00
- Summary generated
- 3 Nov 2026, 11:01
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy