> This time, we used the free/consumer versions of ChatGPT, Copilot, Perplexity and Gemini.
IOW, they tested ChatGPT twice (Copilot uses ChatGPT's models) and didn't test Grok (or others).