Yesterday my wife lost the picture on the TV in her bedroom and called DirecTV Customer Support on the phone. The call went on for quite a while (as such calls do) with the support person having my wife check various settings, changing HDMI port, look up serial number, etc.
As I listened I was impressed with how good this support lady was. Patient, polite (“thanks for confirming”), knowledgable. Maybe the best support call I’ve heard which I mentioned to my wife.
“It’s an AI,” my wife replied.
I was speechless (rare occurrence). I’ve been playing with AI for three years and have heard countless examples of AI “voices” and I’ve always been able to recognize them as such. I was completely fooled by the DirecTV AI.
“Are you sure?” I asked my wife. “Yes. ’She’ said so at least three times during the call.”
Still skeptical, I went to Claude for confirmation:
Voice AI for customer support has gotten good enough that many people can’t tell — that’s actually one of the selling points companies look for. A few things make it convincing:
- Natural-sounding TTS— modern text-to-speech (services like ElevenLabs, or the voice layers built into platforms like Google’s Contact Center AI, Amazon Connect, etc.) captures pacing, intonation, and small imperfections that used to be a dead giveaway of “robot voice.”
- Patience is actually easier for AI — it doesn’t get tired, frustrated, or rushed, so infinite calm is a natural byproduct, not a performance.
- Scripted-but-flexible responses — large phone-support call trees have always been semi-scripted; an LLM just handles the branching more fluidly and can respond to unexpected phrasing.
- DIRECTV specifically — AT&T (DIRECTV’s former parent) and DIRECTV have both publicly discussed using AI/virtual agents in their support lines, so it’s plausible your wife’s call routed through one, especially for a first-tier troubleshooting step before escalating to a human.
Tells that usually surface eventually: it struggles with a genuinely novel or emotional situation, it can’t access account-specific systems a human rep would casually reference, or the call abruptly transfers once the issue gets complex. If your wife is unsure, she could ask directly — some companies require disclosure now, some don’t.
Am I concerned that I can no longer tell when I’m speaking with a human or an AI? Not really. In my experience, humans are really shitty at this job. Bored/apathetic if not just plain rude. Slow. Everyone knows the drill.