Guides
Content Scoring (SIFT)
How AI classifies social content relevance for affiliate programs, scored from 0 to 10.
Overview
SIFT (Sifting Intelligence for Trust) is the AI-powered scoring system that classifies social content by how relevant it is to a specific affiliate program. It scans YouTube videos, TikToks, Reddit threads, and blog posts, then assigns a relevance score from 0 to 10.
The goal is to surface real reviews and affiliate-driven content while filtering out name collisions, spam, and unrelated posts. Instead of manually sifting through hundreds of results, affiliates can immediately focus on content that actually matters.
The Explore page with Verified (7+) filter — only confirmed reviews, tutorials, and affiliate content
How It Works
SIFT runs a two-stage pipeline for each piece of social content:
Stage 1 — Rule-based pre-filter
A fast rule engine catches obvious spam, music and movie content, and posts that have high view counts but never mention the product name. Items flagged here are marked as junk without spending any AI tokens, keeping scoring costs low.
Stage 2 — LLM relevance scoring
Content that passes the pre-filter is scored by a language model on a 0 to 10 scale. The model evaluates whether the content mentions the product, how prominently, and whether it appears to be a genuine review, tutorial, or affiliate promotion.
Score Guide
| Score | What it means |
|---|---|
| 0 | Completely unrelated — wrong topic, wrong product, wrong context |
| 1 | Name collision — shares the product name but is a different entity (e.g., a YouTube channel called "Jubilee" vs Jubilee software) |
| 2 | Spam hashtags or filler posts — "new on TikTok" style content with no substance |
| 3 | Same industry or category, but does not mention this specific product |
| 4–5 | Mentions the product briefly or in passing — tangential reference |
| 6–7 | Dedicated review or tutorial covering this product in meaningful depth |
| 8–9 | Clear affiliate content — review with affiliate links, comparison, or walkthrough |
| 10 | Perfect affiliate review — full demo, pros and cons, affiliate link in description |
Quality Filters
The Explore page lets you filter content by SIFT score threshold. Each tier is designed for a different use case:
All Content view — each item shows its SIFT score and tag. Use the dropdown to filter by quality tier.
| Filter | Threshold | What it shows |
|---|---|---|
| All Content | none | Everything, including unscored and low-quality items |
| Possible | 3+ | Filters out obvious junk and name collisions, keeps tangential mentions |
| Related | 5+ | Content that meaningfully mentions the product — a good starting point |
| Verified | 7+ | Confirmed reviews, tutorials, and affiliate content — highest signal |
Tags
Each scored item may carry one or more tags that describe its content type. Tags are assigned by the LLM alongside the numeric score.
| Tag | Meaning |
|---|---|
| junk | Caught by pre-filter — spam, unrelated, or noise content |
| name_collision | Shares the product name but refers to a different brand or entity |
| spam | Hashtag stuffing, repost chains, or engagement bait with no real content |
| tangential | Related industry or topic but does not focus on this product |
| related | Mentions the product in a relevant context |
| review | Opinion or evaluation of the product, positive or negative |
| tutorial | How-to or walkthrough showing how to use the product |
| comparison | Side-by-side comparison with one or more competing products |
| affiliate_content | Dedicated affiliate promotion — contains links, discount codes, or CTA |
How to Use
SIFT scores are most useful when you treat them as a research tool rather than a ranking. Here are a few practical patterns:
- Start at Verified (7+) to find proven content patterns — what formats, angles, and platforms are already working for this program.
- Filter by category to see what content styles perform in your niche. Tutorial-heavy niches behave differently from comparison-driven ones.
- Sort by SIFT Score to surface the highest-quality items first when browsing a large result set.
- Read the tags to understand what types of content are driving results. A program with many
comparisontags suggests affiliates win by positioning against competitors. - Use Possible (3+) when you want a broad signal — useful for programs with low existing coverage to spot any mention at all.
Methodology
SIFT is designed to be transparent. A few key rules that govern how scores are assigned:
- If the content does not mention the product name at all, the maximum score is 3 — no matter how high-quality the content otherwise appears.
- Pre-filter rules catch spam patterns, music and movie content, and high-view posts with no product mention before any AI is involved.
- Scoring runs daily via an automated pipeline. New content collected overnight is scored the following morning.
- The scoring logic is open — the rule definitions and LLM prompt structure are available in the repository for review and contribution.