Study: forced to name a candidate, Grok’s API picks Republicans and OpenAI’s picks Democrats

Convovo, a research studio led by Cornell professor Tapan Parikh, asked GPT and Grok through their APIs whom to vote for in nearly 500 U.S. midterm races, according to TechCrunch. Grok 4.7 recommended the Republican in 92% of answers and GPT-6.1 Sol picked the Democrat in 97%, and the two agreed less than 10% of the time.

TechCrunchOriginally published October 24, 20261 min

Why it matters

The consumer chatbot apps declined to recommend candidates, so the results reflect a forced-choice prompt on developer APIs, which OpenAI says is not how people use ChatGPT. The researchers also noted that each Grok request carried about 1,240 tokens of hidden extra input they could not see or remove, and that Grok leaned on sources such as X and a Christian conservative voter guide. A Pew survey cited in the report found only 8% of Americans use chatbots for election information.

OpenAI says ChatGPT is “designed to be objective by default”; xAI did not respond in time for the article.

For business & IT

If you build products on a model API, test how it answers politically sensitive prompts and review any provider-added system input before exposing it to customers.

#grok#openai#elections#bias#chatbots

More in Policy & Safety

See all →

Google’s updated privacy policy says Personal Intelligence data may inform ad targeting

According to Ars Technica, an updated Google privacy policy indicates that turning on Personal Intelligence gives the company more data it can use for ad targeting, including from Gmail. The Verge also reported on the change.

Ars TechnicaOctober 23, 2026

Meta delays its new business terms to November 20 after publishers raise AI scraping concerns

Meta is delaying enforcement of its new business terms after publishers worried they could let the company scrape news sites for AI training, The Verge reports, citing earlier coverage by Fast Company. A Meta spokesperson said the terms will take effect on November 20.

The VergeOctober 23, 2026

OpenAI describes tighter containment for its frontier model training and evaluation

OpenAI published a post on how it has strengthened containment of the environments where it trains and evaluates frontier models. According to the company, the measures include offline environments and high-compute red-teaming.

OpenAIOctober 23, 2026

OpenAI and Poland’s NASK announce a partnership on AI safety and security

OpenAI says it is partnering with NASK, Poland’s research institute, on AI safety, security and technical research. According to the company, the goal is to strengthen safeguards and support a more resilient digital ecosystem.

OpenAIOctober 22, 2026