Clear about our limits

Evidence first.
Certainty, never assumed.

ScamClues helps you understand a suspicious message. It does not prove that a person, website, attachment, or request is genuine.

What we look at

The selected original’s sender, reply-to, requested action, language, visible content, link destinations, and available static attachment metadata. Email authentication headers preserved in an uploaded message are historical claims; we do not describe them as independently verified.

Four possible assessments

• Dangerous: strong scam indicators in the supplied evidence.
• Suspicious: meaningful warning signs that call for independent verification.
• No strong warning signs: we did not establish strong scam clues. This is never permission to pay, sign in, download, or share sensitive information without verification.
• Unclear: insufficient, contradictory, unsupported, or unavailable evidence. We need a clearer original or another trusted way to verify.

Why you won’t see “94% safe”

An AI-generated confidence number is not a measured probability. We do not publish an accuracy percentage until independent labels and held-out testing support it. We track false-safe errors, unclear results, evidence formats, and changes between analyzer versions. Testing is ongoing; no overall accuracy claim is currently available.

What we don’t do

We do not open suspicious links, fetch their pages, execute attachments, conduct malware sandbox analysis, call supplied numbers, or provide human review. A link-only submission has limited context. Image-only attachments inside email, QR codes, unusual forwarding formats, and very long messages may require a clearer submission.

When analysis is incomplete

We make limited results visible and do not use your monthly allowance for incomplete or Unclear assessments. Never rely on limited checks to approve a sensitive action.