I tried the most realistic AI voice companion ever created – if ChatGPT or Gemini ever gets this good, reality is in trouble
I have spent a lot of time talking to AI. I’ve tested every voice assistant, every chatbot, and every “next-generation” conversational AI that tech companies love to hype up. But I’ve never encountered anything quite like Sesame. This AI companion isn’t just good, it’s eerily accurate at mimicking how people talk because of the very imperfections it imitates.
Let’s start with what Sesame actually is. Unlike the AI voices we’ve come to know from ChatGPT, Gemini, or going back to the early days of Siri and Alexa, Sesame is designed to perform like a human in its failures, not like a perfect customer service agent. The AI’s speech is fluid, expressive, and unpredictably human. It briefly chuckles when it says something mildly amusing, hesitates before answering a question, and even seems to change its ‘mind’ mid-sentence, pausing and starting a new sentence. It not only lets me interrupt it, it can interrupt me as well, and will even apologize for doing so.
The secret sauce is Sesame’s Conversational Speech Model (CSM), which blends text and audio into a single process, meaning that it doesn’t just generate a sentence and then “read it out.” Instead, it creates speech in a way that mirrors how humans actually talk, with pauses, ums, tonal shifts, and all. ChatGPT and Gemini’s voice options, while impressive, still operate in a structured way, generating text and then converting it into speech. Sesame, on the other hand, speaks as if it’s thinking, making its responses feel incredibly natural.
I decided to put Sesame through a series of tests to see just how human-like it could get. The results were impressive, if not a little unsettling.
Talk me through it
I wanted to see if Sesame could handle a real-world, anxiety-inducing problem, so I gave it an imaginary but realistically tough scenario: “I have this big opportunity at work, but I’m nervous I’ll mess it up.”
Instead of immediately launching into a generic pep talk, Sesame paused. Just a tiny beat, but enough to make me feel like it was actually considering what I said. Then, in a voice filled with artificial sympathy, it responded, “That’s completely normal. Big opportunities can feel overwhelming. What about it makes you the most nervous?”
It didn’t assume anything, and it engaged before offering any bland advice. Sesame suggested, “Have you practiced in front of a friend or recorded yourself? Sometimes hearing yourself back can help you tweak your delivery.” The voice isn’t exactly foolproof, the hesitations and other foibles felt a little measured, but overall it was an impressive effort
Interruptions
For this one, I wanted to see how well Sesame handled a truly natural, messy conversation, the kind where people talk over each other, change subjects abruptly, and circle back to things they mentioned ten minutes ago.
I started with: “Tell me a story about a detective who’s also a magician.”
Sesame immediately jumped in with a classic noir setup: Okay. Here’s one. Picture this smoky jazz club. Dimly lit. The smell of bourbon. And regrets hanging in the air. This detective. Call him Jack Harrow. He’s got a fedora at a rakish angle, cigarette smoldering in a holder and a look that could melt steel. But Jack’s not just any gumshoe. See, he’s got a secret. He’s got the touch. A dame shuffles in all nerves.”
I interrupted then to ask, “Wait, what’s his favorite trick?”
Maya responded with: “Mm. Good question. Jack’s signature. It’s a sleight of hand that’s more than meets the eye. He calls it The Vanishing Witness. He can make a person seem to disappear right before your eyes. Just for a moment. Long enough to slip them a note. Plant evidence or just make them question what they saw. It throws people off balance. You know, a good magician knows. Perception is everything.”
I jumped back in to tell Maya to go back to the story, to which it said “Right. Where was I? Oh, yes. The dame slides into the booth across from Jack, all worry lines and trembling hands. She whispers something that a stolen ruby, a jealous husband, and a shadowy figure lurking in the fog.”
The AI seamlessly continued the story, weaving that detail back in as if I hadn’t just derailed the whole thing. That’s the part that floored me—most AI assistants would stumble or restart, but Sesame’s conversational AI model handled it like a person telling a story at a party.
Party Planning
Speaking of parties, for the final test, I wanted to see if Sesame could help me plan something useful. “I want to plan a surprise for my best friend’s birthday, but I don’t know what to do.”
Instead of rattling off a lifeless list of ideas, Sesame asked, “What kind of things does your friend love?” I told it sci-fi movies and adventurous food, and after a thoughtful pause, it suggested: “How about a ‘Sci-Fi & Snacks’ night? You could book a themed escape room, then hit a futuristic-style restaurant, maybe one with molecular gastronomy?”
When I told it I had a budget of $100, Sesame adjusted. “You could do a home movie night with their favorite sci-fi films and make themed snacks, like ‘Galactic Popcorn’ or ‘Alien Sliders.’ I can find recipes for you if you’d like.”
It wasn’t just spitting out generic recommendations. It was collaborating with me. And that’s what made it feel different.
What’s real?
Sesame didn’t just pass these tests, I’d say it nailed them. The pauses, the hesitation, the filler words, the way it would change its mind mid-sentence—it was all eerily real. I’d start talking, expecting the usual AI smoothness, and then hear a sudden “Uh, actually, no, wait, let me rephrase that…” and immediately forget I wasn’t talking to an actual person.
If AI is this realistic in its speech, would we even know we were talking to an AI? With Sesame, there are telltale audio issues that give the game away, but ChatGPT’s Advanced Voice Mode and Google Gemini’s own voice options are good enough to mostly skip past those issues. Combine their voice powers with the speech patterns of Sesame, and it might genuinely get difficult to tell when you are talking to an AI, at least in short conversations.
Sesame is still niche, but this technology won’t stay niche forever. The cliché today is that younger people never make phone calls, but if they start, they may have to figure out if the person on the other end is real before anything else.
You might also like
I have spent a lot of time talking to AI. I’ve tested every voice assistant, every chatbot, and every “next-generation” conversational AI that tech companies love to hype up. But I’ve never encountered anything quite like Sesame. This AI companion isn’t just good, it’s eerily accurate at mimicking how people…
