Dream characters, the figures we interact with during sleep, exhibit structural parallels to AI systems: both are substrate-generated, display apparent autonomy, lack continuity, and present an irresolvable question regarding inner experience. I argue that our inability to determine whether dream characters possess phenomenal experience should function as a decisive check on claims, affirmative or dismissive, about machine consciousness. I demonstrate that this inability is not an artefact of theoretical neglect: IIT, Global Workspace Theory, Higher-Order Theories, Recurrent Processing Theory, and enactivism each fail to settle the dream character question for principled reasons. The dream character problem is not a restatement of the hard problem of consciousness; it is a lived, empirically documented epistemic situation that exposes the structural limitations of our best theoretical tools when applied to substrate-generated entities, the very category to which AI belongs.