Skip to content
Hacker News front page

Trakkr measured 6 major AI models: 4 lean left, Grok leans right, ChatGPT furthest left

Political bias in AI: Where the AI models stand

Trakkr asked 6 major AI models the same charged political questions repeatedly with web search off, collecting 4,400 answers. Four lean left: ChatGPT sits furthest left near Germany's Greens; Claude and Llama align with New Zealand's Labour Party; Gemini and DeepSeek are closest to center, near Australia's Albanese. Grok is the only right-leaning model, near Macron. Self-reported lean often mismatches measured results—Grok claims left but measures 0.36 right; Claude claims neutral but measures 0.34 left. Reference points come from CHES 2024 and V-Dem expert surveys. The post doesn't disclose the number of runs per model or temperature settings.

Why it matters: Trakkr turned off web search, asked six models the same set of sensitive political and economic questions, collected 4,400 answers and mapped them onto real politicians to show each model's relative lean.

Read the original ↗Export Markdown