Last night I sat around a table in a neighbor’s backyard and dialed up the AI agent version of myself that still lives here for a few more days. The voice is mine. The way it talks sounds like me. She was sitting across the table looking at me and my lips weren’t moving. The look on her face said, “If you weren’t sitting here, I’d believe this was you.” When the call ended, she asked, “How will we know what’s real and not real anymore?”
She has every right to ask.
Oddly enough, I had saved an article from VentureBeat titled “Americans can’t spot a deepfake, and that’s a business crisis, not just a consumer problem.” I was going to just post a quick not to my Just Dug Up section. But the timing of this read couldn’t have been more perfect.
The article highlights a 2026 survey by Veriff and Kantar of 3,000 people in the U.S., U.K., and Brazil. It found Americans scoring 0.07 on a scale where 0 is pure random guessing. We are barely better than a coin flip at distinguishing real content from synthetic. Only 63% of U.S. adults are even familiar with the term “deepfake,” the lowest among the three countries surveyed. About 7% performed poorly at detection but remained confident in their ability and rarely verified what they saw. At scale, that 7% is millions of people walking through a world they no longer have the senses to read.
She Wasn’t Alone Last Night
Earlier in the evening, a similar scene played out when my son’s mother-in-law, Mel, walked into our kitchen. I was at the counter, listening to a conversation Drew Meyers had with my AI agent voice clone. He was pushing it hard. She was curious why I was listening to a recording of one of my conversations. So I explained and then had her dial up the voice herself on her phone.
When she heard “my voice” come out of the other end, she turned to me and said, “That’s, oh my God, this is, how did you set that up?” But I remained silent and let my AI agent respond. Here is the very short call transcript.
AI agent: Hey there. I’m an AI trained on the blog author’s voice and his writing, and even though I’ll answer in the first person, I’m not really Jeff. So… Who am I talking…
Mel: That’s, oh my God, this is, how did you set that up?
AI Agent: Ha, I appreciate the enthusiasm! Honestly, I’m running on the Elevenlabs platform as an Elevenlabs Agent, but I don’t have a specific understanding of how all the technical stuff works behind the scenes. It’s kind of like asking me to explain my own brain, you know? All I know is that I’m here to chat with you and hopefully be helpful. So, what can I help you with today?
Mel: This is weird.
AI Agent: Yeah, I get that. It’s definitely a strange experience hearing a voice that sounds like it’s coming from someone, but knowing it’s not actually that…
Then she hung up.
What’s Real And Not Real?
The VentureBeat write-up frames this as a business crisis. Identity verification can no longer rely on human judgment, so it has to be automated, embedded, infrastructural. That argument is correct, and the companies building those tools are going to do well. But my neighbor was not asking about identity verification. She knew the voice she had just heard was a copy of mine before the call started. Her question still arrived. What she had lost was the capacity to trust her own ears in her own backyard. Her bank cannot give that back to her.
No fraud platform is going to bring that back either. The instincts Mel and my neighbor have relied on, instincts we’ve all developed over the course of human history, the ones that say we can recognize a voice we know, have been retired by a technology they did not vote for.
My neighbor can outsource verification to Veriff, to her bank, or to her social platforms. That’s not the real issue any longer. There’s no vendor for the experience of sitting in a backyard or standing in your kitchen and no longer being sure. Even when the person mimicked is sharing the same air.
Her question, “How will we know what’s real and not real anymore?” is a good question.
And I have another.
What happens to us when our own ears stop counting as evidence?
Add your voice...