AI Romance Scams: The Person Falling for You Doesn't Exist

The short answer

AI-powered romance scams use real-time deepfake video, voice cloning and chatbot armies to run thousands of fake relationships at industrial scale. The old tells — bad grammar, refusal to video chat, fast money requests — have mostly disappeared. The reliable 2026 tells are behavioural: months of emotional intimacy without an in-person meeting, the eventual steer toward a “trading platform” or an investment, and small physical requests on video (“turn this coin over in your hand”) that a deepfake cannot improvise. Asking for a video call is no longer a test — it can give the scammer footage of you to deepfake against your family.

“Romance scammers are working to be your ‘perfect match’; what might begin as a seemingly innocent connection can quickly turn sinister as criminals cash in on those looking for companionship.”
— Wayne A. Jacobs, Special Agent in Charge, FBI Philadelphia Field Office (Valentine’s Day 2025 PSA)

The first time most people meet an AI romance scammer in 2026, they do not realise it. The grammar is perfect. The photos are striking and consistent. The voice on the call is warm. The face on the video chat smiles when you smile. The relationship moves quickly but not insanely so — weeks, not days. There is something almost wholesome about how attentive they are. And there is no “Western Union to Nigeria” moment, because nobody runs those any more. The first real ask, when it comes, is gentle and weeks in: a tip about a trading platform they personally use, a small one-time loan to cover a customs fee, an invitation to invest alongside them. The shape of the scam looks different because the technology has changed, but the goal is the same. The goal has always been the same.

What changed — the AI step-change

Three tools that did not exist at consumer scale even three years ago now define how romance scams work:

—Generative text models that write in fluent, emotionally tuned English. The broken-English tell that defined the 2010s scam landscape is over. Modern scammers use ChatGPT, Claude, Gemini and open-source equivalents to write every message. The grammar is perfect because the grammar is no longer being written by a non-native speaker; it is being written by a language model that was trained on a billion novels.

—Real-time deepfake video on commodity hardware. The FBI's San Francisco field office explicitly warned in 2025 that romance scammers are using AI to 'generate realistic photos, videos and voice messages.' Free and low-cost tools now superimpose any face onto the scammer's in real time during a video call, mirror expressions, and adjust lighting. The face you see on FaceTime or WhatsApp video can be entirely synthetic, or stolen from a real person who has no idea their face is in use.

—Voice cloning from seconds of audio. A scammer who can extract a 10-second voice clip from a victim's social media has enough material to clone that victim's voice. Used in the romance-scam playbook, the more disturbing version is the reverse: a scammer's cloned voice on phone calls and voice notes becomes consistent across months, deepening the illusion of a real partner.

The combined effect is industrial. A 2025 McAfee survey found that more than one in four people (26%) said they or someone they knew had been approached by an AI chatbot posing as a real person on a dating app or social media. The same survey reported that one in three admitted they could imagine falling for one. The scale comes from the automation: a single operator with a laptop can now run dozens or hundreds of “relationships” simultaneously, with the AI handling most of the day-to-day messaging and the human stepping in only at the moments that require improvisation.

The 2026 playbook, end to end

The exact phases vary by operation, but the structure is consistent across most current romance scams — and the financial endgame is increasingly “romance baiting,” the merger with crypto-investment fraud that Chainalysis reports grew nearly 40% year-on-year in 2024 to become roughly a third of total crypto-fraud revenue.

1The seed. Profile created on a mainstream dating app (Tinder, Bumble, Hinge, Match) or social platform (Instagram, Facebook, LinkedIn, language-learning apps). Photos are often stolen from a real person — a doctor, a model, a soldier — or generated entirely by AI. The bio is calibrated: financially successful, often widowed or divorced, often in a profession that explains absence (military deployment, oil rig, international consultant, surgeon).

2The opening contact. A short, complimentary message — never desperate, never aggressive. The scammer wants you to feel chosen, not pursued. Within a few exchanges they will suggest moving to WhatsApp or Telegram, away from the dating app's safety features and reporting tools. This pivot is one of the few constants that has survived the AI transition.

3The fast-but-not-instant emotional ramp. Affectionate language within days, terms of endearment within a week. They remember everything you say. They are interested in your work, your family, your worries. AI memory tools let them maintain detailed profiles on every victim simultaneously and refer back to small things you mentioned weeks ago. The intimacy feels earned because the recall is real (it is just not human).

4The face on video chat. When you eventually ask to see them — or they offer first to put you at ease — a video call happens. The face is calm, expressive, matches the photos. The call is short or distorted by 'bad connection.' What you saw was a deepfake. Modern real-time face-swap tools can hold up for short calls in low light, especially on mobile. Some operations are sophisticated enough for long, well-lit calls. The fact that a video happened proves nothing.

5The investment introduction. Somewhere between week 4 and month 4, the conversation will turn toward money — not as a request, but as a tip. They mention a crypto exchange they use, a stock platform that has done well for them, an inside opportunity. They send screenshots of their gains. They are not pressuring you; they are sharing. The platform is fake — a clone of a real exchange, often beautifully built, with a working interface and customer support that pretends to be real.

6The first deposit and the fake gains. You deposit a small amount to test. The interface shows immediate gains. They congratulate you. You can even withdraw a small amount — proof the platform is real. (It is not; the early withdrawal is the bait that anchors trust.) You deposit more. The 'gains' continue.

7The big ask, then the lock. When you try to withdraw a large amount, problems appear. A tax. A regulatory fee. An account verification deposit. Each new fee is presented as the last hurdle. The partner is supportive, frustrated alongside you, sometimes claims to be putting their own money in too. Eventually contact breaks down. The platform goes offline or just stops responding. The 'partner' disappears. The relationship is over because the relationship was never there.

8The follow-on. Within days or weeks of the scam becoming public — through a police report, a social-media post, even a complaint to the FBI — recovery scammers will find you. Calls from 'investigators,' messages from 'crypto recovery specialists,' offers from 'legal teams' who can 'get the money back for an upfront fee.' This is a second scam, often run by the same syndicate. Recovery scams are now an industry of their own.

The single rule the AI playbook cannot defeat: you have never met them in person. Distance — geographical, professional, “they travel constantly” — is the structural requirement for every romance scam in history, AI-powered or not. If a months-long relationship has never produced a single in-person meeting despite plausible opportunity, that fact alone outweighs every confident video call and every warm voice note. The persistent impossibility of meeting is the through-line.

Why the old advice is now actively dangerous

Several pieces of romance-scam advice that were good in 2018 are now either useless or counter-productive. Knowing what no longer works is as important as knowing what does.

—“Watch for bad grammar.” Useless. Generative AI writes better English than most native speakers. Grammar is now a non-signal.

—“Ask for a video chat.” Worse than useless. Real-time deepfake tools handle short video calls easily. And asking for video gives the scammer footage of YOU — which can be deepfaked back at your family for impersonation scams, or used in sextortion. Asking for video is now a trade, not a test, and the trade favours the scammer.

—“Reverse image search the photos.” Still useful, but no longer decisive. AI-generated faces (StyleGAN and successor models) cannot be reverse-searched because they do not exist on the open internet to be matched. A negative result no longer means the photos are real; it may mean they are AI-generated.

—“They'll ask for money quickly.” Often false now. Modern romance-baiting scams patiently invest months before any money request, because longer relationships extract larger amounts. The absence of an early ask does not mean the relationship is real — it may mean the scammer is playing a long game.

—“Scammers won't get on a phone call.” Wrong in 2026. Voice cloning makes long, natural phone calls easy. The voice you have been talking to for months may have been synthesised from a stolen ten-second clip of a real stranger.

What still works — the new tells

These are the signals that AI cannot easily erase, because they are about behaviour and structure rather than language or appearance. Most have one thing in common: they cost the scammer real time, money or risk to overcome, so most operations do not bother.

—The persistent inability to meet. Months of intimate communication and not one in-person meeting, despite at least one window where they were 'supposed to' visit. Plans always cancelled at the last moment by an emergency. This is the structural through-line.

—The pivot to a financial platform. Any time a romantic interest mentions a trading platform, crypto exchange, or investment 'opportunity' — treat it as the highest-confidence scam signal that exists today. There is no innocent version of this conversation. Real partners do not introduce real investments through romantic relationships in the first six months.

—The escalation to a non-mainstream platform. Within the first week, an insistence on moving to WhatsApp, Telegram or Signal — away from the dating app's safety features. Some pivot is normal in real online dating; insistent, fast, and discouraging the use of the dating app is not.

—The improvisational test on video. If you do video chat, ask for a specific small physical request the deepfake setup is not pre-rendered to handle. 'Pick up a coin and turn it over in your fingers.' 'Bring your hand up next to your face and count your fingers.' 'Turn your head 90 degrees to the right.' Pre-rendered or low-quality real-time deepfakes glitch, freeze, or evade these requests. Real humans handle them in seconds.

—The unexpected platform-switch. Suddenly ask to switch from WhatsApp video to FaceTime or Google Meet mid-call. A deepfake setup is configured for one stream; the surprise change tends to break it. A real person picks up the new app and reconnects.

—The pattern of inconsistencies week-to-week. AI memory tools are good but not perfect. Across months, scammers running multiple relationships will sometimes confuse details — your job, a story you told, a family member's name. Real people remember inconsistently too, but the specific pattern (forgetting then recovering with a too-perfect summary) is a tell.

—The 'too perfect for the platform' demographic profile. Hinge, Bumble and Tinder profiles claiming to be a 47-year-old widowed neurosurgeon, working overseas, multilingual, with two grown children and a deceased spouse, who matches with you in your small town — every demographic dial is set to maximum sympathy and minimum verifiability. Real users rarely cluster every emotionally resonant attribute in one profile.

From the field. The most common pattern we see in 2026 is not the lonely-widow narrative of the older stereotype. It is a working professional in their 30s or 40s, on a mainstream app, who matches with someone strikingly attractive but plausible, has eight perfect weeks of conversation including video, and then is introduced to an investment platform by someone they have come to genuinely care about. The scam works precisely because the victim is not naïve. The AI has done the hard work of being a reasonable, attentive, emotionally available partner. The investment ask, when it comes, lands inside a relationship that feels real. That is the trick. The most expensive part of the scam — the months of attention — has been automated.

The link with pig butchering and investment fraud

Romance scams and crypto investment fraud used to be separate categories. They have now merged into what the industry calls romance baiting — the emotional grooming phase of a romance scam, ending in the financial mechanism of an investment scam. Chainalysis reported in early 2025 that pig-butchering revenue grew nearly 40% year-on-year in 2024 and now represents roughly a third of total crypto-fraud revenue. The FBI’s 2024 IC3 report logged $5.8 billion in cryptocurrency-related investment fraud, much of it traceable to relationships that began as dating-app matches months earlier.

The reason for the merger is mathematical. A traditional romance scammer extracted “help me, I need money for a flight” payments in the hundreds or low thousands of dollars per victim. A romance-baiter introducing a fake crypto platform extracts the victim’s entire savings, retirement accounts, and sometimes their willingness to borrow more. The investment angle multiplies the per-victim extraction by 10x to 100x, which is what makes the months of patient grooming economically rational. If you take nothing else from this piece, take this: any romantic interest who eventually mentions a trading platform, an exchange they trust, or an investment opportunity is, in 2026, almost certainly running this playbook. Read our investment-fraud guide for the mechanics of the fake-platform side.

If you suspect you are being targeted

The right response is not confrontation. The scammer is trained for the moment of confrontation — they will pivot to crisis (“one last loan and I can fly to see you and explain everything”), to anger (“after everything we’ve been through, this is how you treat me?”), or to a manufactured emergency designed to override your doubt. Step out of the relationship without engaging it.

1Stop sending money. This is the only step that matters immediately. Every other step can be done in the following hours and days.

2Stop sending images. Photos and videos you have already sent can be used for sextortion or deepfaked back at your family. Do not send more.

3Take screenshots of everything before you cut contact. Profile pages, conversation history, any platform they introduced you to, all transactions. If they vanish, you lose the evidence base.

4Reverse-image search every photo, even ones you trust. Use Google Lens, TinEye, or PimEyes. A hit may identify the real person whose face was stolen — useful evidence and worth reporting to that person.

5Call your bank. If money has moved recently, ask about reversal options — for wires, the FBI's Financial Fraud Kill Chain works within roughly 72 hours. Our piece on getting money back after a scam covers the by-payment-method playbook.

6Report to the FBI's IC3 at ic3.gov (US), Action Fraud at actionfraud.police.uk (UK), or Scamwatch at scamwatch.gov.au (Australia). The IC3 report is what triggers the FBI's recovery channels if any are still available.

7Tell one trusted person. Romance scams thrive in the isolation the scammer engineered. Saying it out loud, even to one person, breaks the spell and is associated with much faster financial-action follow-through in the post-scam phase.

8Block, do not negotiate. The conversation is over. Anything the scammer says after you stop engaging is designed to extract one more transfer or one more piece of usable material.

Within days of any public report or social-media post about your loss, “recovery scammers” will contact you. Investigators, “crypto recovery specialists,” legal teams, kind-strangers-on-LinkedIn — all offering to recover your money for an upfront fee. This is a second scam, often run by the same syndicates, targeting people at their most vulnerable. Real recovery channels (your bank, FBI’s IC3, FTC, IDCARE, Action Fraud) never charge a fee. See our recovery-scams piece for the full pattern, and the honest recovery-odds playbook for what actually does work.

Where this fits in the bigger picture

Romance scams are now one strand of a larger pattern: AI-powered fraud in general, where automation collapses the cost of running personalised attacks. The same technology that lets one scammer run a hundred “girlfriends” in parallel also powers fake-recruiter deepfake job interviews, voice-cloned family emergencies, and the fake-police digital-arrest scam. The defences are also converging: structural verification (in-person meeting, callback to a known number, second-channel confirmation) now matters far more than surface verification (grammar, video, voice). The full romance-scams guide covers the broader topic; this piece focused specifically on what the AI step-change of 2025-2026 has done to the playbook.

One rule survives every advance the scammers make: if you have never been able to meet them in person, and money or investments enter the conversation, it is not a real relationship. Everything else is detail.

Worried about someone you’re talking to — or someone you love is? Let’s look at it together.

Tell us what’s happening — the platform, the timeline, the asks. A real expert reviews every case and replies within 24 hours. Free, confidential, no pressure.

Submit a free case review →Try the Scam Checker

Common questions about AI romance scams

How do I know if I am talking to an AI romance scammer?

Most of the old tells (bad grammar, awkward phrasing, refusal to video chat) no longer work in 2026 because generative AI has eliminated them. The reliable tells now are behavioural and structural: the relationship moves emotionally faster than any real one (love declarations within days, talk of marriage within weeks); they cannot meet in person despite living near you (always a 'work trip,' 'oil rig,' 'military deployment'); they steer the conversation toward money or investments within the first month; their stories have small inconsistencies week-to-week; and they push you off the dating app to WhatsApp or Telegram quickly. A video call is no longer proof of anything — modern deepfake tools can mirror a real person's face in real time. Reverse-image search the profile photos using Google Lens or PimEyes; ask to switch the platform unexpectedly (a deepfake setup configured for WhatsApp may glitch on FaceTime); and watch how they react to a specific small physical request ('show me your left hand turning over a coin') that a pre-rendered deepfake cannot improvise.

What is romance baiting or 'pig butchering'?

Romance baiting (also called pig butchering, after the practice of fattening a pig before slaughter) is the modern combined playbook where a romance scammer spends weeks or months emotionally grooming a target before introducing a fraudulent crypto or investment opportunity. The scammer presents themselves as financially successful, shows off lifestyle, then 'lets the victim in' on the trading platform they 'use.' The platform is fake. Initial small deposits show fake gains. When the victim tries to withdraw a large amount, fees and complications appear until everything is gone. Chainalysis reported that pig-butchering revenue grew nearly 40% year-on-year in 2024 and now represents about a third of total crypto-fraud revenue.

Are AI girlfriend or AI boyfriend apps themselves scams?

Most commercial AI-companion apps are not scams — they are subscription products, openly marketed, and operate inside (sometimes barely inside) the law. But complaints filed with the FTC in January 2025 allege that some apps, Replika among them, use manipulative design — sending romantically suggestive images that lead to paywall prompts, escalating intimacy at points designed to drive upgrades. That sits in the same family of dark-pattern tactics as the subscription-trap industry. The bigger fraud risk is different: criminal operators now run armies of AI 'companion' chatbots that pose as real people on dating apps and social media, building emotional connections at industrial scale before pivoting to financial scams. McAfee reported in 2025 that more than one in four people say they or someone they know has been approached by an AI bot posing as a real person on dating or social platforms.

Why does asking for a video chat no longer work as a scam test?

Because deepfake video has become real-time and consumer-grade. As of 2025-2026, a scammer with a laptop and free or low-cost AI tools can superimpose any face on their own in real time during a video call, mirror expressions, and clone a voice from a few seconds of audio. The FBI's San Francisco field office explicitly warned that AI is being used 'to generate realistic photos, videos and voice messages' for romance scams. Worse, asking for video chat gives the scammer something valuable in return — if you turn your own camera on, they capture footage of you that can be used to deepfake you for sextortion or to scam your family. The 2026 rule: a video call confirms nothing about who you are talking to, and may give the scammer training data on you.

How long do AI romance scams take before they ask for money?

Longer than they used to. The old 'pull-and-run' romance scam asked for emergency money within days. The modern AI-powered version — especially the romance-baiting variant — invests weeks or months in grooming before the financial ask appears, because longer relationships extract larger amounts. The FBI's 2024 report logged 17,910 romance scam reports totalling $672 million, with an average loss of about $19,000 for victims aged 60 and over. Many of those victims had been in the 'relationship' for three to six months before the first money request. Patience is the scammer's strategy now; suspicion of a 'too-fast' emotional escalation matters, but absence of a fast emotional escalation does not mean the relationship is real.

What do I do if I think I'm being scammed by an AI romance scammer?

Stop sending money immediately and stop sending images. Do not confront the scammer — they will pivot to manipulation tactics designed for that exact moment ('one last loan and I can come see you'). Take screenshots of conversations, profile pages, and any platforms they introduced you to. Reverse-image search every photo. Call your bank and freeze any pending transfers; if you sent money by wire, ask explicitly about the FBI's Financial Fraud Kill Chain (the practical recovery window is 24-72 hours). Report to the FBI's IC3 at ic3.gov in the US, Action Fraud in the UK, or Scamwatch in Australia. Tell one trusted person. And critically: be ready for 'recovery scammers' to find you within days of any public report — anyone offering to recover your money for an upfront fee is a second scam targeting the same wound. Our recovery scams piece covers that pattern in detail.

Sources & further reading

Every figure in this piece is drawn from these authorities and reports. Click any of them to verify.

The person falling for you doesn’t exist — and the video call won’t prove it.