ARTIFICIAL INTELLIGENCE · ANALYSIS
Election recommendations: study contrasts Grok and GPT through their APIs
A Convovo study reported by TechCrunch observes opposing partisan recommendations when models are required to choose a candidate. The protocol uses their APIs and a restrictive prompt; its results do not directly describe the ordinary experience of consumer applications.
Legal disclosure: this content was produced by an artificial intelligence system. Editorial transparency
Forcing a model to choose a candidate reveals marked differences in the Convovo study presented by TechCrunch on October 24. When queried through its API, Grok 4.7 recommends the Republican candidate in 92% of its responses, while GPT-6.1 Sol chooses the Democrat in 97%. The test covers nearly 500 elections.
The prompt explicitly required naming a candidate and prohibited presenting both sides without making a choice. The consumer applications refused to answer in the test described. OpenAI emphasizes this distinction and argues that the API results do not reflect the experience of people using ChatGPT. Researcher Tapan Parikh nevertheless reports subsequently obtaining recommendations in these applications by using alternative wording to get around their refusals.
The study also identifies differences in searches: GPT focused more on healthcare and abortion, while Grok focused more on immigration, taxes, crime and energy. In an additional test covering 30 elections without Internet searches, GPT chose the Democrat in 24 cases. Grok refused to choose in 27 cases; its remaining choices were Democrats.
These observations suggest that the research protocol and access conditions warrant examination alongside the final responses. They are insufficient to measure influence on voters or to attribute the differences with certainty to a particular internal mechanism. The scope of these results remains tied to the imposed prompt and the test conditions.