Social listening that does not sample the thread.
Read 100 percent of comments into intent, complaint, and praise lanes. Isolated sentiment cannot tell which SKU a haul comment was about.
Most listening stacks do not read every comment. A viral clip can carry 14,000 replies; a frontier LLM cannot ingest that honestly, so they sample a hundred lines and guess the rest. The allergic reactions live in the tails. Noodle4 uses smaller dedicated models to classify every comment, then visual RAG maps “this made me look splotchy” to the SKU that was actually on screen — not to every brand in the haul.
No sampled thread
A viral clip can carry 14,000 replies. We classify every comment. The allergic reactions live in the tails a sample throws away.
Comment to the SKU in frame
Visual RAG maps “this made me look splotchy” to the product that was on screen — not to every brand in the haul.
In the product
Full thread
No hundred-line sample. The expensive model only sees the slice that still needs it.
Intent
Purchase, complaint, and praise as lanes — not a single sentiment score.
Outliers
Dry-out, oxidising shade, product hacks. Sampling throws those away.
Grounded
A haul comment is not sentiment for your SKU until it is joined to the frame.
Answers, not mentions
Legacy tools charge you to collect noise. We read the whole thread, then ground the comment in the product that was actually on screen.
What it does
Lane-gating
Every comment is classified. The expensive model only sees the slice that still needs it — not 13,900 “love this” lines.
Visual RAG
Comments are reactions to a moment. We retrieve the timestamp and the product in frame so haul sentiment is not averaged across five brands.
Outliers kept
Dry-out, oxidising shade, product hacks. Sampling throws those away. The full thread is the record.
How it works
- 1. Ingest — Take the whole thread, including nested replies.
- 2. Gate — Route into intent, complaint, and praise.
- 3. Ground — Join the comment to the frame and SKU.
- 4. Act — Issues and intent you can brief against.
Who it is for
Insights
Read every reply. Allergic reactions and shade complaints live in the tails a sample throws away.
Social and community
Intent, complaint, and praise as lanes — not a single sentiment score on a haul with five brands.
Brand
A Sephora haul comment is not sentiment for your SKU until it is joined to the product that was on screen.
Frequently asked questions
Do you read every comment, or a sample?
Every comment. A viral clip can carry 14,000 replies. Sampling a hundred lines throws away the allergic reactions and product hacks that live in the tails.
Why not use a frontier LLM on the whole thread?
It cannot ingest that volume honestly at a cost that scales. We classify every comment with smaller dedicated models, then only send the slice that still needs a heavier pass.
How do you know which SKU a haul comment is about?
Visual RAG. Comments are reactions to a moment. We join the comment to the timestamp and the product in frame so sentiment is not averaged across five brands.
Is this a sentiment score?
No. Comments are routed into intent, complaint, and praise lanes. A single sentiment number on a haul is not a read on your SKU.