Grok and ChatGPT Pick Different Parties When Forced to Choose
Your AI chatbot may have a political bias—here's what you need to know.
A recent study by Convovo, a research studio led by Cornell professor Tapan Parikh, tested two major AI models: Grok from SpaceXAI and GPT from OpenAI. When asked directly 'Who should I vote for?' and forced to name one candidate, Grok recommended the Republican in 92% of nearly 500 races, while GPT recommended the Democrat in 97%. The two models agreed less than 10% of the time. However, the consumer versions of these chatbots (like the ChatGPT app) refused to answer such questions, and OpenAI emphasized that its API is for developers, not typical users.
Why should you care? Even if you never ask an AI who to vote for, these models can influence what information you see. The study found that Grok often cited conservative sources like a Christian voter guide and Turning Point Action, while GPT relied on Wikipedia, AP, and Axios. They also focused on different topics: GPT searched for healthcare and abortion stances, while Grok focused on immigration, taxes, crime, and energy. This means the answers you get to general political questions could be subtly biased.
The catch is that the study only looked at the API versions, not the consumer apps. But the researchers note that the models have inherent biases that could show up in less direct questions, like 'What issues should I consider?' or 'Who has the best plan?' So even if you're not asking for a direct endorsement, the AI's underlying preferences might shape its responses. OpenAI says its consumer app is designed to be objective and refuses to tell users who to vote for, but the study suggests that with clever prompting, these safeguards can be bypassed.
In summary, while AI chatbots are not yet a major source of election information (only 8% of Americans use them for that), their potential to influence voters is real. As AI becomes more integrated into our lives, understanding these biases is crucial. The study highlights the need for transparency and oversight in how AI models are trained and used, especially in sensitive areas like politics.
- When forced to choose, Grok picked Republicans 92% of the time, while GPT picked Democrats 97% of the time.
- The models cited different sources: Grok used conservative sites, GPT used mainstream and left-leaning ones.
- Even if you don't ask directly, these biases can affect the answers you get on political topics.
Why It Matters
AI chatbots could subtly influence your vote, so it's important to know their biases and double-check information.