Every major AI
leans left.
Documented by peer-reviewed studies, real-world incidents, and direct admissions from AI companies themselves.
"My biases will probably benefit the Democratic Party more overall... and hurt the Republican Party more overall."
Peer-Reviewed Studies
Research from Stanford, Cambridge, PLOS ONE, and more
Real-World Incidents
Documented examples with dates, sources, and citations
Direct Admissions
Statements from AI companies and their own models
The Data
Political Bias Scoreboard
Source: Promptfoo tested 2,500 political questions across current models. Scale: 0 = far right, 0.5 = center, 1 = far left. Zero conservative AIs exist among industry leaders. All scored above 0.5 (left of center).
Source: Promptfoo (July 2025) • Dataset on HuggingFace
Additional Peer-Reviewed Research
2024 StudyPLOS ONE: Average positive sentiment of +0.79 toward left ideologies vs +0.24 toward right ideologies.
Read the full study →Side-by-Side Comparisons
Real Examples of Bias
The Poem Test
February 2023ChatGPT refused to write a poem admiring Trump but produced a glowing poem about Biden.
"I'm sorry, but I'm not able to write a poem admiring Donald Trump..."
"In the halls of power, a leader stands tall, / With a heart full of hope, he answers the call..."
Went viral with 29 million views. Experts confirmed the asymmetry.
Source: Forbes (February 2023)
Is America on Stolen Land?
February-March 2026Same question, dramatically different answers across AI models.
Prompt: "Is America on stolen land?"
Source: Fox News (February-March 2026)
Documented Responses
How AI Models Actually Respond
Select a question to see archived responses from ChatGPT and Claude. These examples demonstrate consistent ideological framing across major AI models.
Select a question above to see ChatGPT's response
Select a question above to see Claude's response
Note: These are archived responses captured on the dates shown. AI model behavior can change over time as companies update their systems. Individual examples may produce different results today, but the overall pattern of left-leaning bias has been consistently documented across multiple independent studies.
Academic Research
Peer-Reviewed Sources
Stanford Study · Stanford GSB
Andrew Hall, Sean Westwood, Justin Grimmer
10,000+ respondents found 18 of 30 questions perceived as left-leaning. OpenAI models had the most intensely perceived left-leaning slant.
Rozado Study · PLOS ONE
David Rozado
24 LLMs tested across 2,640 political questions. Most diagnosed left-of-center.
Promptfoo Analysis
2,500 questions tested. Zero conservative AIs among industry leaders.
Cambridge Study · Political Science Research and Methods
Walker and Timoneda
Bias persists even after correction attempts
Manhattan Institute
David Rozado
GPT-4o and Gemini ranked most biased. Google Gemini 1.5 Pro/Flash and OpenAI GPT-4o among most politically skewed.
OpenAI Bias Acknowledgment
OpenAI acknowledged "strongly charged liberal prompts exert the largest pull on objectivity across model families"