In a recent live stream, the channel Grow with AI put several leading AI chatbots to the test by feeding them identical persona prompts and observing how well each model maintained its assigned character.
The experiment covered ChatGPT, Claude, Gemini, and Grok, among others. The host analyzed each chatbot's consistency, creativity, tone adherence, and the point at which it broke character. Early observations suggest that while some models stick closely to the given persona, others drift more quickly, especially when faced with complex or contradictory instructions.
"If you're into prompt engineering or AI research, this one's for you," the host noted.
The session, which streamed live on April 25, 2026, has drawn attention from the AI community and racked up dozens of views within its first week. The video also sparked a small discussion in the comments, with viewers sharing their own experiences and requesting resources like the host's free ebook on persona prompts.
This hands-on comparison highlights both the power and the limitations of current AI in understanding and executing nuanced human instructions—a crucial insight for developers and power users alike.