Every "YouTube automation" post shows you a revenue screenshot and no methodology. Here's the opposite: small, honest numbers, and what they told me.
The setup
I built three brand channels — one each for Korean, English, Japanese — and pushed the first batch of auto-generated Shorts. I deliberately did not dump everything at once. A brand-new channel with zero videos that suddenly uploads 20 clips reads as spam and trips the platform's authenticity heuristics, so I published five to start and staggered the rest.
The numbers (≈4 hours after upload)
| Video | Views | Avg view duration | Note |
|---|---|---|---|
| Daily-fortune | 243 | 0:19 | exceeds clip length → looped, best retention |
| Compatibility | 225 | 0:03 | early drop-off |
| MBTI-match | 206 | 0:05 | early drop-off |
| Salary calc | 3 | — | barely reached |
| Tarot | 2 | — | barely reached |
Subscribers over that window: 0 → 2.
The observed reach — values, not projections.
The diagnosis
Two things fell out of this immediately:
- Reach was fine for the fortune/fun clips (200+ views cold), and terrible for the utility ones (calculator, tarot got single digits). The niche decides whether the algorithm even tries you.
- Retention only held where the payoff was in the video. The daily-fortune clip showed the actual result on screen and got looped past its own length. The others said "get your answer in the app" and lost people in 3–5 seconds. That's the whole finding: in-video payoff beats "answer's in the app." Deferring the payoff to a click is where the watch-time leaks.
Platform gotchas worth knowing
- Shorts descriptions/comments don't pass clicks (since mid-2023). Your only funnel out of a Short is the channel profile link — long-form still carries clickable links.
- The upload cap I hit was a daily video limit on unverified channels, not an API quota. Phone verification raises it.
- OAuth threw a 403 until I added myself as a test user; and it's easy to authorize the wrong (personal) channel instead of the brand one.
- The analytics API lags ~1 day, so same-day judgments come from Studio's realtime view only.
The honest part
243 views and 2 subscribers is not a headline. But it's real, and it pointed at something I could act on: rebuild the format so the result lives inside the video, and stop wasting cold reach on utility niches that the algorithm won't push. A fake "$10k/month" screenshot would have told me nothing. Five small honest numbers redirected the entire build.
If you automate this, publish a tiny batch first and read the retention curve, not the view count. The curve tells you what to change; the view count just tells you the algorithm noticed.