"...we have found strong evidence that xAI, DeepSeek, Anthropic, and OpenAI models tend to discuss controversies from their respective companies in a differentially positive way compared to others. Meanwhile, we found no such evidence for Alibaba, Meta, and Google."

(The magic chart is Figure 2 on page 5.)