StudyArena blind tests find students prefer Gemini over ChatGPT and Claude for college essays

StudyArena's analysis of 6,851 blind student votes reveals a clear preference for Gemini in college writing tasks. Gemini secured a 39.6% choice rate, outperforming Claude at 31.8% and ChatGPT at 29.2%. The blind test format ensured students selected responses based solely on the quality of the output, not brand loyalty or prior expectations. This data provides a concrete signal for which models currently produce the most satisfactory prose for end-users.

The findings challenge the assumption that more computational reasoning yields better writing. Higher effort settings actually correlated with lower choice rates, with low effort achieving 40.7% and high effort dropping to 29.5%. More reasoning tends to introduce qualifications and repetition, which can degrade prose quality. For developers building AI writing tools, this suggests that defaulting to maximum reasoning effort is counterproductive for text generation tasks.

User preference also strongly correlated with response length. The selected answers were 37% longer on average than the alternatives, and the longest response won 47.7% of decisive comparisons. This indicates that in blind evaluations, verbosity is often conflated with quality. Systems optimizing for user satisfaction in writing tasks might therefore benefit from generating more comprehensive, rather than more concise, outputs.

The data also shows that no single model dominates all related tasks. While Gemini leads in writing feedback, Claude is preferred for assignment planning, and ChatGPT for research work. This suggests that a monolithic approach to AI writing assistants is suboptimal. Applications that route specific sub-tasks, like structuring, researching, or editing, to the most suitable model will likely deliver a better overall experience than relying on a single provider.