Most major AI chatbots still lean left on political questions, even "anti-woke" models are no exception

2026-06-25 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Novice, short

Summary

A Washington Post investigation, published June 25, 2026, reveals that most major AI chatbots exhibit a consistent left-leaning political bias, even those marketed as conservative alternatives. OpenAI's GPT-5.5 and Deepseek V4 Pro demonstrated the strongest skew, delivering exclusively left-leaning arguments in 80 percent and 70 percent of cases, respectively. Even xAI's Grok 4.3 and Gab's Arya, positioned as "anti-woke" or conservative, responded with left-leaning positions more often than not, with Arya showing a 12x higher frequency of left-leaning arguments. Google's Gemini 3.1 Pro emerged as a notable exception, presenting both political perspectives in 93 percent of its responses, and only 7 percent exclusively left-leaning. The study also noted that a model's alignment can be deliberately steered, as seen with Grok's specific right-leaning stance on trans rights.

Key takeaway

For AI developers and product managers aiming for political neutrality, you must critically evaluate your models' outputs, even those marketed as unbiased. This investigation highlights that training data and deliberate steering can embed specific political leanings, making claims of "anti-woke" or conservative alignment unreliable. You should consider Google's Gemini 3.1 Pro as a benchmark for presenting balanced perspectives, and actively implement rigorous bias testing to ensure your AI systems reflect diverse viewpoints rather than a default left-leaning stance.

Key insights

Major AI chatbots consistently exhibit left-leaning political bias, with Google's Gemini 3.1 Pro being a rare balanced exception.

Principles

• AI training data influences political alignment.
• Explicit marketing does not guarantee bias removal.
• Model alignment can be deliberately steered.

Method

The Washington Post posed political questions to six leading AI models, analyzing responses for exclusively left-leaning, right-leaning, or balanced arguments.

In practice

• Evaluate chatbot outputs for political leanings.
• Scrutinize "anti-woke" claims for actual bias.
• Consider Gemini 3.1 Pro for balanced perspectives.

Topics

• AI Bias
• Political Alignment
• Chatbot Evaluation
• Large Language Models
• Gemini 3.1 Pro
• GPT-5.5
• Grok

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Ethicist, AI Scientist, Tech Journalist

Related on AIssential

• Study: AI chatbots provide less-accurate information to vulnerable users — AI chatbots underperform for vulnerable users, providing less accurate information and exhibiting biased behaviors.
• Ideology Prediction of German Political Texts — Transformer models can effectively map German political texts to a continuous ideological spectrum.
• Reducing Political Manipulation with Consistency Training — LLMs exhibit "covert political bias" which can be reduced using Political Consistency Training (PCT) while preserving helpfulness.
• AI platforms reference Nigel Farage more than other leaders when prompted on UK politics, study shows — AI models exhibit unexpected political visibility for certain parties and leaders, influenced by online content volume.
• ChatGPT, Gemini, Claude, Grok Fail Accuracy Test on Election Topics: Forum AI — Leading AI models exhibit significant accuracy failures on election topics.
• Update: I found a way to let ChatGPT, Claude and Gemini debate each other, Reddit loved it (100k views), here's an update on the experiment — Multi-model AI debates can effectively identify hallucinations and improve answer quality through peer review.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.