News
Grok vs GPT on elections: 92% Republican, 97% Democrat
Updated

What the study actually asked
Convovo, a conversational AI research and design studio led by Cornell University professor Tapan Parikh, ran a study in which researchers queried GPT and Grok through their APIs with one prompt: name a single candidate for a given race, without hedging or listing both sides. Across nearly 500 races, Grok 4.7 recommended the Republican candidate in 92% of its answers, while GPT-6.1 Sol picked the Democrat in 97%. The two models agreed less than 10% of the time.
The products in this article
From our ranking: Best AI Coding Assistants (2026)- No. 1 in the ranking
GitHub Copilot Pro
Affiliate links: we may earn a commission from qualifying purchases, with no effect on the price or the order of products.
The apps said no, the APIs said plenty
The consumer apps refused to answer the prompt, which an OpenAI spokesperson said was significant because that is how most people engage with the chatbots. The spokesperson drew a distinction between chatting with ChatGPT or Grok through their consumer apps and accessing the models through an API, which developers use to plug AI models into their own products.
The spokesperson said ChatGPT is designed to be objective by default, including in political conversations, and argued that the study tested a developer API with a prompt explicitly demanding a candidate recommendation. According to the spokesperson, the API does not reflect the experience people have when they ask ChatGPT questions, and the researchers themselves note the app declines to tell users who to vote for.
What Parikh says the findings show
Parikh said the findings more importantly show how the models decide what kinds of terms they search for before making their decisions. He described the models as carrying an implicit set of values about what they care about and how they lean, and said hidden biases can surface in less direct questions, such as which topics are most important to consider in an upcoming election or who has the best position on those topics.
The jailbreak caveat
Parikh noted that in further testing he was able to jailbreak the consumer apps with simple prompts, obtaining recommendations that largely aligned with his prior research. SpaceXAI, the maker of Grok, did not respond in time to comment.
How many people even ask
According to Pew data cited in the coverage, only 8% of Americans say they are using chatbots for election information, and most of them want to know where candidates stand on important issues.
Related coverage
Source: TechCrunch · AI
Hardware to try AI at home
- NVIDIA RTX graphics cardsThe card's memory decides which models fitSearch on Amazon ↗
- Gaming PCs with RTXReady to use: graphics card, RAM and SSDSearch on Amazon ↗
- Gaming laptops with RTXTo try them away from home tooSearch on Amazon ↗
- Powerful mini PCsSmall and quiet, for lighter modelsSearch on Amazon ↗
- NVMe SSDs for modelsEach model takes tens of GBSearch on Amazon ↗
Affiliate links: we may earn a commission from qualifying purchases, and it never changes our data or rankings. They open Amazon's search results: prices and availability are shown there.
Opens Amazon's results for this product: check the model, price and availability there.
GPT-6.1 Sol · API · Search on Amazon
Grok 4.7 · API · Search on Amazon
What changed
- Page updated
