The problem isn't "no script" — it's "feels like marketing copy"
When teams work on Xiaohongshu (RED) content, the core issue often isn't a lack of script — it's that the script reads too much like a script. The product points are all there, the information is complete, but it sounds more like a product description than a real user sharing a personal experience. On Xiaohongshu, this distinction matters significantly.
Case background: skincare content brief
This case came from a skincare content brief. The team already had a reference video they wanted to learn from, plus an initial voiceover draft. But the existing copy felt stiff — packed with efficacy claims and missing the lived-experience perspective that Xiaohongshu audiences respond to. The goal wasn't to invent a new topic, but to rewrite the existing material to fit Xiaohongshu's native voice.
Deconstruct first, AI Review second — not generate-first
For this kind of work, ShenBi AI doesn't jump straight to generation. The pipeline begins with reference video deconstruction and AI Review. The system first identifies what actually makes the reference video work: why the opening lands, where the pain point appears, which expressions are worth borrowing as method, and which would feel too "salesy" if copied directly.
Three key insights from deconstruction
After deconstruction, three findings stood out. First: the reference video's real effectiveness wasn't in the product names themselves but in the sequence — describe the problem, then the felt experience, then the solution. Second: the original draft was full of generic phrasing like "suits all skin types" and "noticeably effective" — flat language that comes across as empty. Third: even though all the selling points were present, the original lacked the "why I tried this" and "the moment I noticed the difference" framing that authentic Xiaohongshu sharing typically uses.
From "product description" to "lived experience"
Based on these insights, ShenBi AI didn't just expand the product-description direction. It rewrote the script toward lived-experience framing: starting from a specific skin concern instead of leading with product info; breaking the middle into smaller, more digestible segments instead of stacking efficacy claims; and ending with a "who this is for / who this isn't for" judgment instead of a hard push. The result reads like a real user sharing rather than a templated recommendation voiceover.
Final selection: the most natural-sounding version
The system output multiple candidate versions. The team didn't pick the one with the most information — they picked the one that sounded most natural and most aligned with real spoken-Xiaohongshu pacing. The final draft kept the core selling points but shifted the overall feel from "product description" to "personal experience sharing".
What this case is really about
The most valuable part of this case isn't that AI quickly produced a script. It's that the system first extracted what actually worked in the reference video, then pulled the original template-feeling copy a step closer to authentic Xiaohongshu voice. For platforms like Xiaohongshu, this step matters more than adding a few more selling points.
If you already have reference videos but don't know how to turn them into scripts that sound like a real person sharing — this kind of pipeline is more reliable than asking AI to generate from scratch.