AI Science Tracker
EN

Country names and query language shifted AI answers in different ways

University of Copenhagen; Aalborg University Copenhagen; Bocconi University · 22 Oct 2026 · humanities

Discovery

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.

Open in the full catalog for filters, charts and related discoveries. Browse humanities discoveries.

Related catalog entries

Links use documented relationships where available, plus shared catalog context. A shared topic or model family does not establish a scientific dependency.