How it works
A channel that learns from every batch
Faceless Theory runs your Shorts channel the way a scientist runs an experiment. Each batch tests one change. The winner joins your channel's style. Then it tests the next thing.
What happens each day
A batch is a set of Shorts made and posted together, usually six. Before anything is made, the app picks one thing to test, such as how the first sentence is written. Half the batch gets version A, half gets version B, and everything else stays the same.
The batch moves through four stages. Each has its own AI agent and its own chat, so you can see and change every step.
- Review & test. Reads the numbers from the last batch, decides which side won, and proposes the next test.
- Plan. Writes a spec for every video: the topic, the script and which side of the test it's on.
- Produce. Builds each video in parallel: voice, images, word-by-word captions and motion.
- Post. Schedules the batch to YouTube, alternating A and B.
You can approve each stage, or switch on autopilot and only look at the results.
The rules every test follows
These come from running the method by hand on our own channel. Most of them exist because we got a test wrong without them.
- One change per batch. If two things change at once, you can't tell which one did it.
- At least three videos per side, four recommended. In our first batch, each side had one video that took off and one that got almost nothing. One video proves nothing.
- Alternate A and B when posting. A, B, A, B, so the time of day can't explain the result.
- Every video records every setting. Before calling a result, list anything else that differed between the sides.
- Judge on rates, not raw views. Since March 2025, a Short counts a view the moment it starts playing.
- Compare medians. One runaway video shouldn't decide a test.
- Give it time. Wait until every video has 300 views, or 72 hours have passed.
- "No clear winner" is a real result. Keep the current style and test something else. Come back to it later.
- Your ideas become tests. Ask for a change and the app suggests it as the next test, rather than quietly switching.
The numbers it judges by
| Number | What it means | Best for judging |
|---|---|---|
| Stayed to watch | The share of people who watched instead of swiping away | Hook, first frame, topic |
| Average % watched | How much of the video people saw. Above 100% means rewatches | Script, length, pacing, ending |
| Engaged views | Views where people kept watching | Overall reach, once the style settles |
| Subscribers per 1,000 views | How many viewers became subscribers | Topic, ending, channel fit |
What it can test
The hook, the first frame, the script's shape, the length, the voice, the ending, the image style and the topic. For a new channel it starts with the hook and the first frame, because those decide whether anyone stays past the first second.
Each winning result becomes a line in your channel's style, with a note of the batch that proved it. Over time the style becomes a record of what your audience actually watches.
Why it stays on the right side of YouTube's rules
YouTube can stop paying channels whose videos look mass-produced from one template. A channel that tests a new change every batch doesn't look like that, because every batch is different on purpose and each change is a real editorial decision.