AI romance scams: why a face, a voice, and even a video call no longer prove your match is real
For years the advice was simple — do a reverse image search and ask for a video call. Generative AI has quietly broken both checks. Here is what changed, and the one test that still works.
A romance scam used to be exposed by two free tricks: a reverse image search on the photos, and “let’s hop on a video call.” For years that was the whole defense, and it worked. Generative AI has quietly broken both. The face can now be invented from nothing, and the call can be staged in real time. This is what changed — and the one test that still holds.
What AI changed
The fraud is the same; the tools on top of it are new. The FBI’s Internet Crime Complaint Center warns that criminals are using generative AI across exactly these schemes — “social engineering, spear phishing, romance schemes, confidence fraud, and investment fraud.” Four old defenses have weakened:
- The fake face has no original. A reverse image search catches a stolen photo because it appears elsewhere under a real person’s name. But the FBI notes criminals now “create realistic images for fictitious social media profiles” — a synthetic face that exists nowhere else. A clean search result no longer means the person is real.
- The conversation runs itself. Criminals “create voluminous fictitious social media profiles” and “create messages to send to victims faster”, reaching “a wider audience with believable content.” Instant, fluent, around-the-clock affection is cheap to produce at scale.
- The voice is borrowed. With a short clip lifted from anywhere online, a scammer can “generate short audio clips containing a loved one’s voice.” A phone call that sounds like a real person proves nothing.
- The video call can be staged. The hardest blow to old advice: the FBI warns criminals now “generate videos for real time video chats.” Real-time deepfakes are still imperfect — more on the tells below — but “ask for a video call” is no longer a clean pass/fail.
The fact to anchor on: a photo, a voice, and even a live video are no longer proof of a real person. Each one can now be manufactured.
The playbook underneath hasn’t changed
Strip away the AI and the structure is the same one the FTC has documented for years. Scammers profess love fast, then keep themselves conveniently unreachable in person: the FTC found the most popular excuse for never meeting is “a faraway military base”, with “offshore oil rig worker” another common (and fake) occupation. The contact usually starts on social media — about 40% of people who lost money said it began there — before the scammer moves the sweet talk to WhatsApp, Google Chat, or Telegram, away from any platform that might flag it.
And the destination never changes. The FTC reported nearly 70,000 romance-scam reports in 2022 with losses of $1.3 billion and a median loss of $4,400. The FBI’s 2025 Internet Crime Report puts confidence and romance fraud at about $929 million in reported losses — fifth among all loss categories — within a record $20.877 billion in total internet-crime losses, up 26% on the year. The tech on the surface is brand new; the place it leads — a request to move money you cannot get back — is identical.
How to spot it
The tells are in the box above, but the pattern is this: a relationship that lives only online + every in-person plan that collapses + an eventual ask to move money = stop and verify. AI makes the persona more convincing; it does not change where the road ends.
If you do get on a video call, the FBI lists what to watch for in AI-generated footage: “distorted hands or feet, unrealistic teeth or eyes, indistinct or irregular faces, unrealistic accessories such as glasses or jewelry, inaccurate shadows, watermarks, lag time, voice matching, and unrealistic movements.” Spotting any of these is a strong warning sign — but their absence is not a clearance.
What to do instead
You cannot out-verify a good fake with another screen-based check, so stop trying to. Lean on the one thing AI cannot touch and the one rule that cannot be deep-faked:
- Meet in person before money. The single test AI cannot stage is a real person in a public place. Until that happens, treat the identity as unconfirmed — however perfect the photos, voice, or video.
- Keep The Pause on Money absolute. Never send money, gift cards, crypto, or account access to someone you have only met online. The FBI’s own guidance is to “never share sensitive information with people you have met only online.” If a relationship has already turned toward an “investment” or trading platform, that is the pig-butchering crypto-romance pattern — read that next.
- Talk to someone before you act. Isolation is part of the method. An outside view is often the thing that breaks the spell.
For the older, photo-based checks and how to use them kindly, see how to spot a fake dating profile and the warning signs of a romance scam. To see how the same deepfake-video trick is used to authorize payments at work, read the deepfake video-call payment scam. Then test your eye on the 60-second quiz or browse more defense moves.
- The relationship lives entirely online, and every plan to meet in person quietly collapses.
- A reverse image search of their photos comes back clean — no other matches — because an AI-generated face has no original to find.
- Replies are instant, fluent, and tireless at any hour, day after day — polished affection that never seems to cost them effort.
- On video, small things are off: a face that does not quite track, lag when they should react, or hands, teeth, jewellery, and shadows that look wrong.
- However the story unfolds, it always arrives at the same place — a request to move money you cannot get back.
- Treat identity as unprovable online. A photo, a voice, and even a live video can all be AI-generated — so none of them, on its own, proves the person is real.
- Make the one test AI cannot fake the deciding one: meet in person, in a public place, before any money, gift, or "investment" is on the table.
- Never send money, gift cards, crypto, or account access to someone you have only met online — no emergency, deadline, or promise changes that rule.
- If you do a live video check, watch for deepfake tells (lag, a face that does not track, off hands or teeth). Treat passing it as "not disproven", never as "proven".
- Talk to someone you trust before you act. Scammers isolate and rush you; an outside view breaks the spell.
For years the advice was simple: do a reverse image search and ask for a video call. I taught people exactly that. The hard part of writing this was admitting both moves now carry an asterisk — AI can manufacture the face and stage the call. What it still cannot manufacture is a person sitting across a table from you, and it cannot turn a wire transfer back around. When the clever tests fail, the boring ones hold.
Sources
Frequently asked
I did a reverse image search and found nothing. Doesn't that mean the photos are real?
Not anymore. A reverse image search works by finding where a stolen photo appears elsewhere online. But AI can generate a brand-new face that exists nowhere else, so a clean result is no longer proof. The FBI notes that criminals now create realistic images for fictitious social media profiles. A no-match result means "not caught", not "verified".
We video chatted and it really looked like them — isn't that proof?
It is much weaker proof than it used to be. The FBI warns that criminals now generate videos for real-time video chats. Real-time deepfakes still often have tells — lag, a face that does not quite track, odd hands, teeth, or shadows — so a glitchy call is a red flag. But a smooth call is not a guarantee. Identity is best confirmed in person.
What's the one rule that still works no matter how good the fakery gets?
In a scam, money only flows one way — away from you, by a method you cannot reverse. AI can fake a face, a voice, and a video, but it cannot change that. If someone you have only met online asks for money, gift cards, or crypto — or wants you to "invest" — it is a scam, every time.
Is it rude to ask someone to prove they're real?
No. You can verify kindly. Suggesting you meet in person in a public place, or simply noticing that every plan to meet falls through, is not an accusation — it is basic safety. A real person who cares about you will understand.