How TrustScore is calculated

The model

Every domain starts at 100 / 100. We then subtract weighted penalties for each risk signal we detect. The final score maps to one of three risk bands:

• 80–100 · Looks Safe
  No major risk signals detected.
• 50–79 · Use Caution
  Some risk signals — verify before sharing info.
• 0–49 · High Risk
  Multiple or strong risk signals detected.

We also publish a confidence score with every TrustScore. It reflects how many of our signals returned real data versus "unknown" for the domain. Low confidence means we couldn't gather much — treat the result as preliminary.

Signals and weights

We don't penalize signals we couldn't measure — "unknown" is never treated as bad.

• Domain age

  −10 to −30 points

  Domain Identity · Source: Public WHOIS records

  Newly registered domains (under 1 year, especially under 30 days) are statistically more likely to host short-lived scams. Brand-new domains lose the most points.

• WHOIS hidden

  −10 points

  Domain Identity · Source: Public WHOIS records

  Many legitimate sites also hide WHOIS for privacy, so this is a small penalty — only meaningful in combination with other signals.

• No HTTPS

  −40 points

  Connection Security · Source: Direct request to the URL

  A site that does not use HTTPS cannot encrypt anything you submit. This is a structural problem regardless of intent.

• Listed in a known threat database

  −50 points

  Reputation · Source: Google Safe Browsing API

  The domain has been independently flagged by Google Safe Browsing for malware, phishing, or unwanted software at the time of the scan.

• AI content patterns (high)

  −25 points

  Content · Source: Google Gemini, on extracted page text

  An LLM detected strong scam patterns in the page text — typically high-pressure urgency, fake giveaways, gift-card payment requests, or impersonation language.

• AI content patterns (medium)

  −15 points

  Content · Source: Google Gemini, on extracted page text

  Some suspicious patterns were detected, but not enough to be confident. Treated as a caution signal, not a strong indicator.

• Brand impersonation

  −30 points

  Brand Spoofing · Source: Lexical match against ~80 commonly impersonated brands

  The domain name closely resembles a well-known brand (e.g. paypa1.com, amaz0n-support.net). False positives are possible for legitimate resellers — disputes are reviewed.

• Very low traffic / popularity

  −10 points (only when known)

  Traffic · Source: Public ranking estimate; skipped when unknown

  Brand-new sites with no traffic at all is one weak indicator. We never penalize a site simply for being unknown to us.

• User reports

  −5 each, capped at −30

  Community · Source: Verified reports submitted by signed-in TrustLayer users

  Signed-in users can report a site (with reason). Reports are weighted but capped — the score cannot collapse from reports alone.

How fact-check assessments work

The Article Fact Checker takes the text you paste and asks an AI model to assess its credibility against publicly available sources. Every assessment is a hedged, opinion-based signal — never a declarative verdict.

Signals we use

• Per-claim source matching (supporting, contradicting, or context).
• Manipulation patterns: clickbait, emotional language, anonymous sourcing, missing dates, fabricated-quote indicators, AI-generated text patterns.
• References in the text to doctored or edited media (text-only — we do not analyze images or video).
• The AI's own confidence score, lowered when public sources are sparse.

What we explicitly do NOT do

• Verify image or video authenticity (no deepfake detection).
• Render verdicts on opinion pieces, op-eds, or satire.
• Render verdicts on claims about identifiable private individuals.
• Score the credibility of named journalists or public figures.
• Replace primary sources, qualified journalism, or expert analysis.

Fact-check verdicts use four hedged labels — Likely Credible, Mixed Signals, Likely Misleading, and Unverifiable. We never use absolutes like "False" or "Fake News". If you believe a fact-check assessment is wrong, file a dispute from the report and we will review it.