The usual answer is "under 60 seconds" or "as long as it needs to be", both of which are true and useless. We have twenty published Shorts between 23 and 50 seconds and their view counts, so here is a more specific answer with its caveats attached.
What our numbers look like
The channel's two best videos are 34 and 30 seconds, at 1,269 and 574 views.
The longest batch we published, six videos running 40 to 50 seconds, sits between 11 and 221. The shortest video on the channel, 23 seconds, took 176, which is above the median but not near the top.
Read naively that says shorter is better. We do not think it says that, for a reason worth explaining, because it is the same reason most content-length advice is worthless.
Why we do not trust our own data here
The long videos were not just longer. They were a different batch, written later, published on a different schedule, after we had changed several other things at once. One earlier batch was suppressed entirely because we published six videos ten minutes apart, which had nothing to do with length and everything to do with timing.
So length is tangled with batch, schedule, and script quality in our data, and we have no way to untangle it with twenty videos. Anyone presenting a clean length-to-views curve from a sample this size is showing you a coincidence.
What we can say is narrower and still useful: we overshot the length by accident, and it made everything harder to compare. We estimated script durations from word counts and got it wrong, so one batch came out ten to fifteen seconds longer than the rest of the channel. That is not a disaster, but it means those videos cannot be compared cleanly against the others on anything, including the experiment we were actually running.
The rule that survives
If you are testing anything at all across videos, hold length roughly constant on purpose. It is a free variable to control and an expensive one to have moving in the background.
For the story format specifically, the practical range is 30 to 40 seconds, and the constraint that sets it is not the algorithm. It is that a petty-drama story has one setup, one escalation and one payoff, and that structure fits in about that much speech. Stretching it to 50 usually means the escalation got padded, and padding is exactly the part people scroll away from.
Turning words into seconds
Since script length is what you actually control, the conversion we measured on our own scripts:
| Script | Finished video |
|---|---|
| 75 to 95 words | 30 to 40 seconds |
| 110 to 130 words | 45 to 50 seconds |
That is for a conversational synthetic voice reading short lines. A faster voice or longer sentences will shift it, so measure your own once and then trust your own number rather than ours.
The practical version: write to about 85 words, check the first finished video against the clock, and adjust the target once. Then stop thinking about length, because it is the least interesting decision in the format.
If you want the part that does matter, it is the first line: we wrote up what a script for this format actually requires, and the full numbers from all twenty videos including the ones that failed.