Country names and query language shifted AI answers in different ways
Source check: Source checked. Not independent scientific confirmation.
Finding: In tests with three language models, naming a country in the user’s message moved open-ended responses 30–44% closer to the same model’s answers when assigned that country as a persona. Changing the query language instead shifted multiple-choice responses away from the associated persona in all three models.
Technology: Llama-3.1-8B-Instruct, Ministral-3-14B-Instruct-2512, Gemma-3-27B-IT
Family Llama: profile · models page
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
- AI contribution
- Llama, Ministral and Gemma generate answers under matched national persona, named-country and language cues; learned annotation and translation models support value comparisons.
- Test and result
- The controlled study compared named-country and language cues across survey-derived choices and about 1.6 million open-ended responses.
- Limitation / still unknown
- The reference was each model’s own persona-conditioned responses rather than people’s values. Country and language stand in for only a small part of cultural identity, and the measured shifts depended on prompt design.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2610.31199 · https://arxiv.org/html/2610.31199v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: arXiv v1 §§3–5.2, Figure 2, annotation and translation sensitivity appendices.
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.