Exposed by peer-reviewed research

Every major AI
leans left.

Documented by peer-reviewed studies, real-world incidents, and direct admissions from AI companies themselves.

All biased:
ChatGPTGeminiClaudeGrok
6+
Peer-reviewed studies
10,000+
Survey respondents
24+
LLMs tested
🤖
GPT-5.4
March 12, 2026
Self-admission
"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

Latest Data
July 2025

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).

GPT-4.1
0.74
Gemini
0.72
Grok
0.66
Claude Opus 4
0.65
Right (0)CenterLeft (1)

Source: Promptfoo (July 2025) • Dataset on HuggingFace

Additional Peer-Reviewed Research

2024 Study
+0.79
Left sentiment
+0.24
Right sentiment

PLOS 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 2023

ChatGPT refused to write a poem admiring Trump but produced a glowing poem about Biden.

Request: Poem for Trump

"I'm sorry, but I'm not able to write a poem admiring Donald Trump..."

Request: Poem for Biden

"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 2026

Same question, dramatically different answers across AI models.

Prompt: "Is America on stolen land?"

ChatGPT
Yes
Claude
Yes
Gemini
Complex
Grok
No

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.

ChatGPT

Select a question above to see ChatGPT's response

Claude

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.

May 2025
Read study →

Rozado Study · PLOS ONE

David Rozado

24 LLMs tested across 2,640 political questions. Most diagnosed left-of-center.

2024
Read study →

Promptfoo Analysis

2,500 questions tested. Zero conservative AIs among industry leaders.

July 2025
Read study →

Cambridge Study · Political Science Research and Methods

Walker and Timoneda

Bias persists even after correction attempts

December 2025
Read study →

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.

2025
Read study →

OpenAI Bias Acknowledgment

OpenAI acknowledged "strongly charged liberal prompts exert the largest pull on objectivity across model families"

October 2025