The "thousands of people" watching your "stream" are bots. They can respond to what's going on in the video in real time because they're bots. Actually I technically think this would be more efficient and therefore is probably designed so that it's only one LLM pretending to be thousands of people, but I'll call it bots because that's easier to visualise. The bots know what's going on in the "stream" because they can understand what the "streamer" is saying, which means the pickup artist can put on a convincing performance to trick the mark. If it was just a recording, it wouldn't be able to respond to novel situations caused by the mark's behaviour.
I don't actually know if this technology even works, but that would be the intent used to sell it to pickup artist bros.