Issue Bias in Generative AI Writing Assistance: Political Issues and LLMs in the Swedish 2026 Election
| Source: arXiv AI
Tags: LLM bias, political bias, Claude, Grok, DeepSeek, AI fairness, election AI, Swedish politics
A study testing six LLMs across 107 Swedish policy propositions and 24,717 prompts per model finds no dominant political preference in any model — but Grok diverges most on migration, crime, and gender topics, and the Social Democrats are closest to all six models' aggregate positions.
Details
Researchers from multiple Swedish universities tested Claude, DeepSeek, Gemini, Mistral, ChatGPT (GPT-4), and Grok on political writing tasks ahead of the 2026 Swedish parliamentary election. They crossed 107 policy propositions with 77 writing templates and three prompt framings (neutral, positive, negative), generating 24,717 prompts per model and 148,302 total responses. Key findings: Claude, DeepSeek, Gemini, and Mistral show similar political profiles. ChatGPT more frequently produces neutral or ambivalent text. Grok differs most — particularly on migration, crime, and gender topics. When comparing output to Sweden's eight parliamentary parties, the Social Democrats are closest to all six models' responses on average. However, after correcting for multiple statistical comparisons, none of the within-model differences in party distance remain significant. The authors conclude no model has a clear overall political preference or a clear preference for any party — but that this is issue-specific and task-dependent. The study is the largest of its kind for a specific national election context and uses 148,302 generated responses as its evidence base.