TL;DR AI hallucinations in analytics are risky because a made-up metric is indistinguishable from a real one — there’s no visual signal that anything is off. Three hidden risks drive nearly every incident: invented metrics, flawed source data that the AI quietly adopts, and incorrect metric definitions applied without any warning. Each of these can be avoided, but only if you know which questions to ask. Seven key questions distinguish AI tools that surface verified answers from those that merely generate believable ones. Databox AI is designed around these seven questions — anchored in your connected data, with traceable outputs, confidence indicators, and a governed metric library. AI is only reliable when it runs on clean, connected, consistently defined data. Ignore that foundation and you simply scale inconsistency across the organization. Introduction A VP of Marketing shares an AI-generated performance report on Monday morning. CAC figures look solid. Trend lines point in a clear direction. The executive summary suggests shifting $200K in budget from paid search to organic content. The CFO agrees. The reallocation is signed off before lunch. Two weeks later, an analyst double-checks one metric against the original source. The figure doesn’t appear anywhere in the connected data. The AI fabricated it from patterns learned during training, displayed it in the same format as every validated metric around it, and no one noticed because there was nothing visually unusual. This is how AI hallucination shows up in analytics. Not blatant nonsense, but believable numbers, formatted just like the real ones, sitting…