The study described in the article is sound (randomised, pre/post attitudes, 600 debates), and outside experts add useful context. But the article blurs findings: AI matched humans overall and only outperformed them with personal data, mainly on low-stakes topics. Debates were short, online and artificial, so real-world election manipulation is speculative. The 64% statistic lacks context, and the paragraph on analytical style and guessing is muddled.