Methodology
How the Legit Score works
An opinion computed from reviews we can see. Every input is shown on each product page.
01 Filter
- Incentivized: reviewer disclosed a freebie/payment, or the text says “received free”, “sponsored”, “#ad”, “use my code”, “affiliate link”, etc.
- Brand-posted: appears on the brand’s own site/channel or is posted by a brand account.
- Creator hype: 5★ with heavy superlatives (“game changer”, “life changing”), exclamation spam, ALL CAPS and little concrete detail.
Filtered reviews stay visible so you can judge for yourself. The current filter is keyword heuristics; with an LLM key it becomes a model classifier.
02 Score four signals
Independence share25%
independent reviews ÷ all reviews
Independent sentiment35%
avg independent ★ (credibility-weighted)
Rating consistency15%
paid-vs-independent gap and spread
Return/defect mentions25%
share mentioning returns, refunds, defects
03 Verdict
80+ Looks legit60–79 Legit, with caveats40–59 Proceed with caution<40 Hype exceeds reality
Under 8 independent reviews = low confidence.
Data sources
- YouTube Data API v3 — not connected
YOUTUBE_API_KEY - Reddit API — not connected
REDDIT_CLIENT_ID, REDDIT_CLIENT_SECRET, REDDIT_USER_AGENT - Web search API (Brave / SerpAPI / Bing) — not connected
SEARCH_API_PROVIDER, SEARCH_API_KEY - LLM (OpenAI-compatible) classifier & summarizer — not connected
LLM_API_KEY, LLM_BASE_URL, LLM_MODEL
This demo runs entirely on fictional SAMPLE data.