Loading page
About this tool
Judge realistic AI-written messages as scam or genuine and get scored on recall, false alarms and balanced accuracy.
The AI Scam Message Awareness Quiz shows you realistic texts, emails, chat messages and call transcripts — including genuine ones — and scores your verdicts using standard classification metrics: recall (the share of scams you caught), specificity (the genuine messages you left alone) and balanced accuracy, the average of the two. Because generative AI removed the broken-grammar tell, the quiz trains the signals that still hold: one-time-code requests, irreversible payment methods, lookalike domains, manufactured deadlines and contact arriving on an unexpected channel. Built for staff security training, older relatives and anyone who wants to test their instincts rather than assume them.
Open AI Scam Message Awareness Quiz on AltFTool — it loads instantly in your browser.
Provide your input — an image, text, or data.
Let the tool analyze or generate the result.
Review, refine, and reuse the output wherever you need it.
A third of the bank is real notifications, so the score punishes crying wolf as well as missing scams.
Every reveal lists the specific tactics in play — urgency, code requests, lookalike domain, secrecy.
Recall, specificity and balanced accuracy are reported separately instead of a single flattering percentage.
You usually cannot, and that is the point — generative text removed the spelling and grammar errors people were trained to look for. Judge the request instead of the prose: what is being asked for, how urgent it claims to be, which payment or verification method it pushes, and whether the channel is one you already use with that organisation.
No. A one-time passcode exists specifically to prove it is you, so any caller or message asking you to read one back is an attack, including callers who already know your name, address and recent transactions. Genuine code messages arrive on their own and typically say the sender will never ask you for the code.
Hang up and call them back on the number you already have for them, or message a second family member to confirm. Voice cloning needs only a few seconds of audio, so a familiar voice is no longer evidence — agreeing a family code word in advance gives you a fast, offline check.
Balanced accuracy is the average of recall (scams correctly flagged) and specificity (genuine messages correctly cleared), which prevents a good-looking score from a lazy strategy. Marking every message as a scam gives 100% recall but 0% specificity, so balanced accuracy lands at 50% — the same as guessing.