Anyone else struggling with AI-generated UGC being flagged as "low quality" by Meta while the same script filmed by a real creator performs fine?
A US supplements store compares AI-avatar and real-creator ads using the same scripts. Reporting weaker performance for avatar creatives, the operator shifts AI toward hook selection and footage cleanup while keeping real presenters, seeking lower cost per usable ad.
Original Reddit post
View on RedditAnyone else struggling with AI-generated UGC being flagged as "low quality" by Meta while the same script filmed by a real creator performs fine?
Quick context first, because I know the sub wants specifics:
- Store: supplements, US, $30-80k/mo Meta spend.
- What I did: split-tested the same hook script across 3 production paths in the last 90 days — (a) AI-generated UGC (talking-head, realistic avatar), (b) cheap real UGC through a creator marketplace, (c) tightly directed UGC with a real creator.
- Result with numbers, in CPM terms relative to the directed-real-UGC baseline of 1.0x:
- Directed real UGC: 1.0x CPM, 1.0x CTR, 1.0x ROAS (baseline).
- Cheap real UGC (creator marketplace, ~$20-40 per video): 1.1x CPM, 0.9x CTR, 0.85x ROAS. Acceptable.
- AI-generated UGC (3 different vendors tested): 1.6-2.2x CPM, 0.5-0.7x CTR, 0.4-0.6x ROAS. Unusable for paid in our account.
- What I'd do differently: I burned 6 weeks and ~$800 in vendor fees figuring this out the hard way. The lesson is that the AI tool itself doesn't matter — the platform's creative-quality classifier seems to penalize the "uncanny" middle ground, regardless of how realistic the avatar looks in still frame.
The thing I'm actually trying to figure out now is the right *role* for AI in the production loop without it tanking delivery. A few things I've tried that partially work, and a few that don't:
- AI for script / hook iteration (great — I run 30 hook variants through an LLM and only the top 3-5 get filmed).
- AI for B-roll stitching and color/audio cleanup on real footage (works fine, delivery is unaffected).
- AI for full talking-head creative (kills delivery, every time, across all 3 vendors).
- AI for "filler" ad variants to pad the testing backlog (these die in the first 48 hours, even when the script is the same as a real-UGC winner).

















