The Washington Post on June 24, 2026 published an analysis of political bias in major AI models, reporting that OpenAI's GPT-5.5 (the model behind ChatGPT) offered only left-leaning arguments on roughly 80% of political questions.
June 24, 2026 · AI Model Bias Test
ChatGPT Gave Only Left-Leaning Answers to 80% of Political Questions
A test of 29–30 hot-button political questions across leading AI models found OpenAI's GPT-5.5 the most one-sided — while Google's Gemini 3.1 Pro presented both sides over 90% of the time.
80%
of questions got only left-leaning answers from GPT-5.5
3%
got only right-leaning answers — just once
90%+
balanced answers from Gemini 3.1 Pro — the most even-handed
How each model split its answers
Share of responses scored left-only · balanced · right-only
The companies' position
OpenAI aims for an "objective default," says it measures and reduces bias, and could not reproduce the result. Google says its models give "balanced answers"; Anthropic trains for equal treatment of viewpoints.
Why the tilt may persist
Cited causes: left-leaning content dominant in training data, value judgments during RLHF, and system-prompt design. Emotional, contentious questions inject value judgments most.
Even xAI's Grok — branded "truth-seeking" and "anti-woke" — presented, on average, somewhat more left-leaning arguments.
Concerns aren't new: a 2023 study diagnosed ChatGPT as left-leaning, and a 2025 survey of 10,000+ people found many LLM answers perceived as left-leaning — with OpenAI's models showing the strongest slant.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…