The specific failure that science communicators describe is not that AI produces bad prose. It's that AI content pipelines start with the study's conclusions and work backward, while science writers start with the question of what is surprising — and those are not the same workflow. A Bluesky user described a study whose unexpected findings were the interesting part, buried by a content specialist with an AI tool who packaged the results without recognizing what distinguished them [1]. That is not a quality problem; it is a sequencing problem. The…