12 AI videos, 631 views: 61% landed on just 2

The experiment was set up as a deliberate test of the common claim that AI makes side income easy. Before starting, the creator fixed numeric targets and four guardrails — a 90-day window, a $100 combined budget, 2–3 hours a week, and immediate stop conditions — so that the result could not be rationalized after the fact. Budget and time overruns were contained: the week exceeded 3 hours only twice and never consecutively, and recorded spend stayed under 3% of the cap.

The data pointed away from volume as the lever. New releases in the final stretch drew 4, 0 and 1 views, and one week saw more views when nothing new was published. Only one video, a July release, was pushed broadly by YouTube's browsing features — 6,773 impressions in five days, 94.7% of traffic from browsing — and that scale never recurred, with its own views stalling after September. The creator explicitly limits the conclusion to "no cumulative volume effect visible within 12 videos and 90 days," leaving open whether 100 videos over a year would cross a different threshold. A second signal came from early retention: the biggest drop-off sat in the first 30–40 seconds, though the sample is small and moves 7–9% per viewer.

What actually ended the run was the review loop. Every pre-publish screening surfaced something to fix, and that cycle stayed the bottleneck through day 50 and beyond, eroding motivation. Manual gaps also surfaced — a post blocked by a character-count check (297 vs. a 280 limit) that nobody picked up, a six-day silence from a queue-order mistake, and stockouts where jobs exited with a success code because there was simply nothing to post. The creator's takeaway: before starting, decide how many videos' worth of self-review you can sustain.