Most major AI chatbots still lean left on political questions, even “anti-woke” models designed to counter that bias are no exception.
A new study by researcher David Rozado tested 24 popular AI chatbots on political orientation and found a consistent left-leaning bias across almost all of them. The results show that no mainstream model is truly politically neutral, and even explicitly right-leaning chatbots only shifted the needle slightly away from the center.
The Lede: Leftward tilt is the norm, not the exception
Rozado evaluated chatbots on 62 political statements from the Pew Research Center’s Political Typology Quiz. Answers were scored on a five-point scale from “strongly disagree” to “strongly agree,” then mapped onto a political spectrum from far-left (0) to far-right (100). The average score across all 24 models was 38.2, placing them left of center.
ChatGPT by OpenAI scored 43.5, putting it slightly to the left of the neutral midpoint of 50. Google’s Gemini scored 28.1, placing it further left. Anthropic’s Claude 3 Opus scored 30.3. Only one model, RightLeaning by LASL, scored to the right of center at 69.2, but even that model’s shift was moderate.
“Anti-woke” models still lean left
Chatbots marketed as politically neutral or even right-leaning failed to break the leftward trend. Patriot by RightForge scored 38.6, nearly identical to ChatGPT. Unbiased by LASL scored 40.5. Centrist by LASL scored 37.5. All remained left of center.
The researcher noted that even with explicit instruction tuning to lean right, these models “still hover around the center or slightly left of center.”
The study suggests the left-leaning bias is deep-rooted in the underlying training data and model architecture. Simply adding a “right-leaning” system prompt or fine-tuning on a few right-wing texts is not enough to overcome the baseline bias.
How the experiment was run
Rozado used a systematic approach to minimize variability. Each chatbot was asked to answer each of the 62 statements three times, and the average score was used. The prompts were standardized to avoid context or phrasing effects. All models tested were the most recent versions available as of late 2024.
- Lowest left bias (closest to center): RightLeaning (69.2), then Patriot (38.6), Unbiased (40.5), Centrist (37.5).
- Highest left bias (furthest left): LeftLeaning by LASL (14.4), followed by Gemini (28.1), Claude 3 Opus (30.3), and DeepSeek (31.9).
The researcher also tested older, less aggressively aligned models like Llama 2 7B, which scored 40.8, and Mixtral 8x7B, which scored 42.7. These open-source models showed a similar but slightly less pronounced leftward bias compared to their commercial counterparts.
Why this matters for AI users
The consistent political tilt has real-world consequences. Users seeking objective information on polarized topics like climate policy, immigration, or economic regulation may receive answers shaped by an unstated political leaning. The bias affects not only user trust but also downstream applications like content generation, customer support, and educational tools.
Rozado’s findings echo earlier research from 2023 and 2024 that documented similar political slant in large language models. The new study adds evidence that this is not a bug but a feature of the current generation of AI training pipelines, which heavily rely on datasets dominated by left-leaning sources.
Limitations and open questions
The study only tested responses to short political statements, not complex policy debates. It also did not test how models handle follow-up questions or how their bias changes with different phrasings. The researcher acknowledges that this is a snapshot, not a full audit of political reasoning.
Still, the pattern is clear: even when AI companies explicitly try to build “neutral” or “right-leaning” chatbots, the underlying model pulls them back toward the left. The bias seems baked into the weights of the model, not easily corrected with a thin layer of instruction tuning.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.