Why a Sephora haul comment is not sentiment for your brand
In a five-brand haul, “this made me look splotchy” is not negative for every product on screen. Isolated comment scoring cannot tell which SKU the viewer meant.
A comment on a haul or GRWM video is not sentiment for every brand in the clip. “This made me look splotchy” has no brand by itself. Basic AI assigns that line to every product in the description, the OCR, and the transcript — so four of five assignments are false. Comments are reactions to a moment. Without the frame, “this,” “that shade,” and “the last one” have no referent. Visual RAG connects the comment to timestamps and on-screen SKUs so the complaint maps to the product that was actually visible, not to the whole haul.
The noisy-video context gap
The error compounds on Sephora hauls, GRWMs, and “what’s in my bag” formats — the formats that dominate beauty and fashion. The more products in frame, the worse a brand-level sentiment rollup becomes, and the more confident the dashboard looks because comment volume is high.
Example: one comment, five brands
Isolated scoring versus mapping the comment to the frame| Comment | Naive assignment | Visual RAG |
|---|
| “This made me look splotchy” | Negative for all five brands | Complaint on the SKU in frame at that timestamp |
| “The last one is my holy grail” | Positive for the video | Praise for the final product shown |
| “That shade oxidised” | Negative brand health | Friction on the named or on-screen shade |
You cannot analyse a comment without the frame
We treated this as a retrieval problem, not a prompt problem. Visual RAG retrieves frames, SKUs, and metadata, then maps the thread to the product the reply chain is about. A splotchy reaction on product three is a complaint for that SKU. It is not a crisis for the other four.
How we know this
These notes come from building Noodle4’s review and listening pipelines, not from a published academic sample. Cost curves, intent splits, and ASR collisions are from internal R&D and client-shaped tests. They will not match every category. We include them because the failure mode is structural — not because one campaign is universal.
Alex Gray is the founder of Noodle4 and a former technical AI product owner. He works with agencies to apply multimodal AI to creator screening, draft approval, and post-publication monitoring.
Frequently asked questions
Why is sentiment wrong on haul and GRWM videos?
Those videos contain multiple brands. A comment like “this made me look splotchy” has no brand by itself. Tools that score comments in isolation attach the complaint to every product in the video.
What is visual RAG for comments?
Visual RAG retrieves the frames, SKUs, and timestamps a comment is reacting to, then maps the complaint or praise to the specific product on screen instead of the whole video.