Methodology
How ScamScan decides
The short version: AI reads the wording, but it never decides the verdict. Every risk level comes from fixed, auditable rules — so the same evidence always produces the same answer.
Step 1 — The AI extracts signals, nothing more
Your pasted message (or attached screenshot) goes once to Anthropic’s Claude, which answers one narrow question: which known scam wording patterns appear in this text? It returns true/false flags with short supporting quotes. It is explicitly forbidden from guessing facts about the world — whether a company exists, whether a person is real, or whether a link is malicious — and it never outputs a risk level.
- Upfront payment or equipment fees
- Cryptocurrency deposits or transfers
- Task-scam structure (top up an account to “earn”)
- Requests to install software or share credentials
- Pressure to move to WhatsApp/Telegram
- Offers with no real interview
- Early requests for bank or identity details
- High pay for a vaguely described role
- Generic mass-send wording
- Unusual urgency or secrecy
- Claims of a senior role (checked against profile activity)
Step 2 — Deterministic checks add independent evidence
Alongside the wording signals, plain code checks the things code checks best: whether contact domains match the company’s official site or imitate it (lookalike detection), how recently domains were registered (public RDAP records), whether links hide behind shorteners or silently redirect somewhere else, whether a URL is on Google’s Safe Browsing blocklist, and how your reported LinkedIn profile observations line up.
Step 3 — Fixed rules assign the verdict
A small set of hard rules decides the risk level. Some evidence is decisive on its own: any request to pay upfront, deposit crypto, move money through your accounts, or install software forces a Dangerous verdict no matter how good everything else looks — a polished profile can’t launder a payment request. Otherwise, strong signals (like a domain mismatch or off-platform pressure) and soft signals (like vague high-pay wording) combine on a fixed scale. If there isn’t enough evidence, we say so instead of guessing.
What ScamScan never does
- Never calls a result “safe” — a clean scan means we found no red flags in what you provided, not that the role is real.
- Never claims a person is or isn't a scammer — it assesses the interaction, not the individual.
- Never stores your message or screenshot — content goes once to Anthropic for classification and is not retained by ScamScan.
- Never scrapes LinkedIn or looks anyone up — profile answers come from what you observed yourself.