Voice control has three stages: hearing the wake word, converting speech to text, and matching that text to a device or routine. A failure at any stage produces the same unhelpful reply.
Naming is the most common problem. Two devices with similar names, or a name that sounds like a common word, makes matching unreliable — and room assignment matters just as much, because 'turn off the light' depends on knowing where you are.
Acoustics come next. Background music, hard surfaces and distance all reduce recognition accuracy, and a speaker placed in a corner hears far worse than one in the open.
Finally, routines silently override individual commands, so a schedule can undo what you just asked for within seconds.