Recently, I have been producing one or two YouTube videos almost every day.
Over the last few days, however, the view counts have been almost absurdly low: one view, two views, and perhaps even those came from me checking the upload myself.
Of course that is disappointing.
These videos are not being thrown together carelessly. I use AI to explore topics, then keep revising the script myself. The thumbnail takes time. Production takes time. The workflow also consumes a fair number of tokens.
What makes the drop harder to understand is that the channel did receive some initial distribution. A few early videos reached more than 90 or 100 views. Those numbers were not large, but they at least suggested that YouTube was willing to test the content with an audience.
Then the views largely disappeared.
Recent video view counts, captured on August 28, 2026.
The hardest part may not be the low view count
Low views are discouraging, but I think the deeper problem is the imbalance between effort and return.
If every video requires finding a topic, writing, recording, editing, and designing a thumbnail, several hours can disappear before the upload is ready. Getting one or two views after all that work feels difficult to justify.
It naturally raises the question: what was all that time for?
My own reaction has been more stable than I expected, largely because much of my production workflow is automated.
The system still needs supervision and adjustment, and it still consumes tokens, but the amount of manual time I put into each video is much lower. When a video goes nowhere, I do not feel the same imbalance between a large personal investment and almost no visible result.
That may be one of the less obvious benefits of automated content production. It does not only make production faster. It also reduces the emotional cost of each failed experiment.
For the first 30 days, I am trying not to decide too early
I have heard several YouTube creators give similar advice: during the first 30 or even 35 days of a new channel, stop checking the numbers constantly.
Keep publishing. Build a meaningful sample. Then look for a direction.
For now, I have decided to follow that advice.
It is too early to decide whether this channel can work. A video may fail because of the topic, the thumbnail, or a script that cannot hold attention. YouTube may also have not found the right audience for the channel yet.
At this point, I cannot distinguish among those explanations.
Instead of letting one or two views reset my mood every day, I would rather give the experiment enough time and enough samples. I will keep publishing for at least 30 days, then look back and decide what appears to work and what should be abandoned.
I am not only building an automated publishing system
I have also been changing the way the scripts are written.
I study other channels' topics, thumbnails, and narrative structures, then ask AI to learn from different approaches. I cannot yet prove that any one approach works, but I am at least not repeating exactly the same process 30 times and hoping for a different result.
I have also built a data-feedback feature.
Through the API, it can collect a video's performance after 24, 48, and 72 hours. Once a video has enough real viewers, I can examine the retention curve as well:
- Do viewers leave within the first 10%, or do they reach 20% or 30%?
- Where does the drop become obvious?
- Did the opening fail, or did the script lose momentum later?
The value of these data is not simply that they label a video as good or bad. They can show the AI where a script lost attention and provide evidence for how the next script should change.
With one or two views, that analysis is not useful yet. The sample is too small for a retention curve to mean anything.
But the system I want to build is becoming clearer:
Find topics that may interest an audience. Study content that has already earned attention. Produce a video. Collect first-party data. Feed the result back into the next production cycle.
This is not one-off content generation. It is a loop that can potentially learn.
What can I control?
I am still willing to continue because many variables in this process can be observed and improved.
Other channels' public view counts can show which topics earned attention.
Their titles and thumbnails can be studied.
Their scripts and narrative structures can be broken down without copying their identity or material.
After my own videos are published, I can record impressions, click-through rate, views, and audience retention, then use those signals to change the next topic and production cycle.
AI can participate in each of these steps, and the workflow may improve as the evidence accumulates.
Some things remain outside my control.
I cannot decide whether YouTube will distribute a new channel, when it will test an upload, or which viewers will see it first. Rewriting a prompt one more time does not solve those uncertainties.
What I can do is improve the controllable parts while accepting that the uncontrollable parts are real.
My understanding of making money on YouTube has changed
When people think about making money on YouTube, attention naturally goes to the outcome: when will the views arrive, when can the channel monetize, and how much can a video earn?
After actually starting the work, I think the early stage is better understood as a search for a viable angle.
Different topics need to be tested. Different scripts need to be tested. Different thumbnails need to be tested. AI can reduce the cost of each attempt, automation can increase the speed of experimentation, and real data can show which direction deserves another round.
None of this means that using AI and automation guarantees a successful channel.
If YouTube never distributes the content, or if the chosen direction has no meaningful audience, even a polished system will not make the experiment work.
But while I can still test new topics, angles, and forms of expression, the experiment is not over.
So yes, I am disappointed when a new video receives one or two views. I am simply not ready to stop.
At this stage, the goal is not to pretend that the channel has already succeeded. It is to complete enough experiments and collect enough evidence to learn whether an AI production-and-feedback system can eventually find a direction that works, and perhaps becomes a sustainable source of revenue.