Pause the conversation
Don't send money, passwords, or one-time codes while you check. Don't follow a link just because it appears in a message.
A MOMENT OF CLARITY
Something feels off?
Take a second look.
Understand the cues in a message.
Then choose a safer next
step.
01 / CHECK
Pause.
Read the evidence.
Verify independently.
Your message
ENGLISH · TEXT
0 / 6,000
Remove personal details first. Links are read as text and never opened.
OR TRY AN EXAMPLE
?
YOUR SECOND LOOK
Clarity starts here.
Your evidence and verification plan
will appear after you
check a message.
01
Read the cues02
Check the source03
Choose a next step
Read the cues02
Check the source03
Choose a next step
KEEP YOUR NEXT STEP SIMPLE
Verify through a
channel you trust.
A convincing message is not proof of identity.
Find your own contact
Use an official app, a saved contact, or a website address you already know. Avoid contact details supplied in the suspicious message.
Ask about the request
Confirm both the identity and the specific action being requested. A familiar name, logo, voice, or writing style can be impersonated.
If you've already acted
Contact the service through its official support channel. If money was sent, contact your payment provider promptly. Change exposed passwords through the official app.
Further guidance:
FTC phishing guide ↗
OPEN, EXPLAINABLE, LOCAL
Evidence over
black-box answers.
A small trained model. A transparent decision process.
103 of 106 spam predictions correct
103 of 126 spam messages detected
20% split · fixed random seed 42
How a second look works
- Text stays in your browserUnicode normalization and English word tokenization.
- Trained language patternsTF-IDF word and phrase features → logistic regression.
- Independent checksRequests for codes, payments, urgency, and link structure.
- Your verification planEvidence and next steps. Never a “safe” verdict.
Know the limits
Trained on the historical UCI SMS Spam Collection. Spam labels are not fraud labels. These results do not establish performance on modern scams, emails, or AI impersonation.
Duplicates were removed before the split. Near-duplicates may remain. Short, unfamiliar, and non-English messages receive a limited-context notice.
No external URL reputation checks, identity verification, or AI-content detection. No claimed reduction in fraud.
Dataset & attribution · CC BY 4.0 ↗