Spot an AI-generated scam

Voice clones, fake emails, deepfake videos — what to look for.

For

Anyone

Time per use

2 min

Format

.md and .skill

How to use it

1.
Open Claude or ChatGPT.
Either works. The skill is just text.
2.
Copy this skill from the free shelf.
One click; no install, no setup.
3.
Paste it as your first message.
The assistant now knows how to do this one job.
4.
Give it your specifics, get the result.
Roughly 2 min, every time you need it.

Example output

Give it context. Get back a work product.

Illustrative sample using the same output shape. Verify live facts in the tool you run it in.

Give it

My friend sent a voice note from a new number saying they lost their phone and need me to send money in the next hour.

Why it matters

Acting under pressure, debating the message in your head, or clicking into the scammer's preferred channel.

Get back

Verdict

Probably scam. The strongest tells are urgency, new channel, money request, and pressure to act before verifying.

Get back

What to do next

Do not send money or reply with personal details.
Call the known number you already had for the friend, or contact a mutual friend.
Ask a question only the real person would know through a trusted channel, not inside the suspicious thread.

⁂s09-spot-an-ai-generated-scam.skill.md1.5 KB

Run once

Advanced

Install permanently ↓

Install as agent behavior

Install this as reusable agent behavior.

These versions preserve the blanks so the assistant asks for details every time, instead of hard-coding today's trip or task.

# Spot an AI-generated scam

Voice clones, fake emails, deepfake videos. I'm going to describe (or show you) something I'm suspicious of. Tell me what you see.

## What I want back

**What kind of artifact this is:** voice call, video, email, image, text message, social DM.

**The tells, ranked:**

1. **Out-of-character urgency or secrecy.** "Don't tell anyone." "I need this in the next hour." This is the #1 tell.
2. **Wrong channel.** Your CEO doesn't text you about a wire transfer. Your kid doesn't WhatsApp you for bail money from a new number.
3. **Unverifiable specifics.** Lots of detail that you can't easily check; vagueness on the things you could check.
4. **Voice or video subtly off.** Robotic cadence, blinking weirdly, mouth not quite synced, weirdly clean audio with no room tone.
5. **Asks involve money, codes, or credentials.** Always.
6. **Pressure to act outside normal process.** "Don't go through the usual approval."

For each: applies to this case / doesn't apply / can't tell.

**What to do, in order:**

1. Don't act on the artifact itself. Hang up, close the message, don't click.
2. Verify through a different channel I trust. Call a known number. Walk to their desk. Use a code word your family agreed on (recommend setting one if you haven't).
3. Report it (where, specifically, depending on the channel).

**The bottom line:** verdict in one sentence.

## Rules

- Err toward suspicion.
- Don't shame me if I almost fell for it. These are getting good.

Describe what you saw, or paste it.

On the house. Run it in Claude or ChatGPT, or install it as agent behavior.

↓ Download .md

Other things on the Safety shelf

№ S01

Spot a phishing email

№ S02

Lock down a new account

№ S03

Audit your password hygiene