@huskistaken has 560K followers on TikTok, 19M total likes, and 4 videos above 10M views. Every single one of them uses the same format.
He films himself in selfie mode, phone in hand, ChatGPT’s voice model running out loud. He talks to it in a deliberately slow, slightly dim-sounding voice, as if he is giving instructions to an expert who works for him. The AI responds with a mildly irritated, confident tone. Then something goes wrong for the AI, and the contrast is the joke.
The hook is always some version of “How AI reacts to…” and the video is always a setup that exposes a gap between what the AI claims it can do and what it actually does.
His best-performing video runs this mechanic perfectly. “How AI reacts to OpenAI CEO reacting to my video” hit 15M views, 17K comments, and 111K bookmarks on April 4.
He shows the AI a clip of Sam Altman saying the voice model cannot track mile times, so he asks the model to track his mile time. The AI agrees. He shows it the clip. The AI says that was a different voice model. Then he tells the AI he just ran a mile and asks how long it took. The AI says 7 minutes. He ran for two seconds. The AI has no idea.
The other three videos above 10M follow the same gotcha logic.
“How AI reacts to me stuck in quicksand” reaches 14.1M views by having him announce an emergency in a completely flat, monotone voice from his bedroom, exposing how the AI responds to crisis prompts it cannot verify.
“How AI reacts to my face filter” hits 10.8M by having the AI describe a face filter that does not exist, producing a backhanded description of his real face.
And “How AI reacts to simple instructions” reaches 10.3M by telling the AI to laugh at everything, then announcing his grandma died, leaving the AI stuck between its instruction and its moderation rules.
Every video is a different scenario. The format never changes.
What makes this replicable for any AI voice app is that the format only requires a voice model with enough personality to feel like a character. The creator’s slow, demanding delivery only works because the AI sounds like it has an opinion. That friction is the whole show.
Any AI companion, voice assistant, or daily tool app could run this exact format by finding the version of the gotcha that fits their product. The misconception does not have to be about ChatGPT. It just has to make the AI sound like it is in over its head.