I expect we're gonna hear more about the cherry-picked data points that aren't atrocious for AI. Like specifically the "help me learn" usage stat (the only one where some AI users are on par or better), and the even more specific "help me learn" usage with AI/media literacy discussed in class stat.
There's probably all kinds of interesting factors involved that had to be ignored because they needed an AI fluff statistic. Like the one fact they mention, that high AI-use and partly non-AI use correllate with curiousity. Kids who want information will use the tools given to them, and those are often LLMs. If your teachers are into AI and discuss it in class, even more so. I expect the "once or twice a year"-users (who bottom out every chart) to include a big bulk of kids that just do not care about education (often coming from unpriviliged households), so they use it to cheat on their assignments and call it a day.
Pretty undeniable though, that the daily users are always worse than the sometimes users. That's probably the habitual delegation of thought kicking in. I could accept that a subgroup of the AI users learns on par or sliiiightly better than non-users on average, but they might be building up a reliance on tooling that will serve them worse and worse once they get past well-trodden school problems. In the same vein they're probably all getting fucked, because noone gives a fuck about good text books & reference materials anymore.