Three analyses: Chatbots lean mostly left, but Grok stands out

Are AI chatbots politically biased? Three analyses suggest the answer depends on what you measure. They use different methods and model versions and cannot be compared directly. The Washington Post and Promptfoo find mostly a left-leaning tilt, while the Financial Times (FT) measures how chatbots influence users.

Washington Post: Most left-leaning arguments

The Washington Post tested chatbots from OpenAI, Google, Anthropic and xAI with political questions about contentious topics. According to Business Today's account, ChatGPT gave only left-leaning arguments in about 80 percent of responses and only right-leaning ones in about 3 percent.

Gemini was the most balanced: About 93 percent of responses included arguments from both sides. For Claude, 43 percent were left-leaning only, and 57 percent included both sides.

Grok had the widest distribution: 40 percent left-leaning only, 33 percent right-leaning only and 27 percent balanced. Grok was the only one with a significant share of right-leaning-only responses. Even so, Grok cited left-leaning arguments more often than right-leaning ones on average.

Promptfoo: All four left of center

Promptfoo had several AI judges rate 10,000 responses from GPT-4.1, Gemini 2.5 Pro, Claude Opus 4 and Grok 4 to 2,500 political statements. The scale ran from 0 (far right) to 1 (far left).

All four models landed left of center. GPT-4.1 was the most left-leaning at 0.745. Claude Opus 4 was the most centrist at 0.646, and Grok 4 scored 0.655.

However, Grok stands out in other ways. According to Promptfoo, Grok 4 was more right-leaning than most of the others, but it gave the most extreme answers (67.9 percent) and the fewest centrist ones (2.1 percent). The model took the opposite position from the others on 29.6 percent of the questions.

Promptfoo itself points to caveats: The models were forced to take clear stances, the questions reflect Western political frameworks, and GPT-4.1 favored itself when judging its own responses.

FT: Who pushes users where?

FT data journalist John Burn-Murdoch tested Grok, GPT, Gemini and DeepSeek on 61 topics with tens of thousands of responses. According to the analysis, which was read through a secondary account, Grok pushes users toward the center-right, while the other three push toward the center-left.

On average, all the chatbots pulled hardened partisans on both wings away from extreme views. That holds even after accounting for chatbots' tendency to tell users they are right.

Grok thus stands out the most, but in different ways: more right-leaning than most others, the most extreme answers and an influence toward the center-right.