For a month, I used AI to run my YouTube channel semi-automatically. The results showed that more than 70 videos were uploaded and the training time was mostly less than five minutes. 10 subscribers and 70 hours of cumulative viewing were recorded.
In terms of numbers alone, the most interesting thing about this experiment is that the more data it builds up, the better the quality of the answers to the AI tools that YouTube provides.The AI, which initially offered only somewhat concise suggestions, began to give contextual advice as the channel’s hysteria accumulated.
The content of this channel itself is not a major material. So I came to this conclusion. If you have only a good idea, you can try enough to make a video with AI. If you only get the planning power, the barriers to the production resources are lower than you think.
In the process of production, we used the open source video editing tool Freecut. Working with this tool, we found that the consistency of the tones between the videos was definitely improving. Voice is working with Qwen, a level that is stable to use.
There is also an interesting point: as the number of videos increases, when a new video is uploaded, the viewing rate of previous videos also rises a little bit. The more content in the channel, the more they attract each other.
It is not yet clear whether the overall viewing pattern has changed during the summer holidays, or whether the algorithm has begun to recognize the channel as low quality.
There’s one thing that’s unlikely: I felt that this combination alone was not enough if it was a channel that I’d like to care about, so I would personally recommend using tools like Seedance or Google Flow. In the end, it was a month when I once again felt that it was more important to choose a tool that suits the channel’s orientation than to use a tool.
Loading comments.