The profile looked genuine. Four photos, each from a slightly different angle, each with the kind of natural lighting and small imperfections that real selfies have. She said she was 31, an architect in St. Petersburg, recently single, and she messaged him first. After two months of daily messages, he agreed to send $1,800 for an emergency dental procedure. By the time he realized the photos showed a person who had never existed, the money was gone and the account had been deleted.
What changed in this type of Russian romance scam is the first step. The photos were not stolen from someone’s Instagram. They were generated by AI. Nobody’s identity was borrowed. A completely synthetic person was built from nothing, and it was convincing enough to sustain a two-month relationship.
The original approach to building fake Russian dating profiles was simple: find an attractive woman on social media, copy her photos, and use them on a dating site. It worked for years. The problem, from a scammer’s perspective, was that reverse image search made those stolen photos findable. Run a profile photo through Google Images or Yandex, and a match might come up, revealing that the “Russian woman” was actually a fitness influencer from Kazan whose real account was easy to find.
AI-generated photos solve this problem entirely. A GAN (generative adversarial network) or a diffusion model like those powering Midjourney or Stable Diffusion can produce a photorealistic human face that has never existed and therefore appears nowhere online. There is no source image to find because there is no source. Traditional catfishing relied on stealing real photos from other people’s social media accounts, but AI-generated photos create unique, original images that have never appeared anywhere online before, created in seconds and designed to look like a completely real person who does not exist.
The practical result is that the reverse image search test, which used to be the first thing many people tried, now frequently returns nothing even when a profile is completely fabricated.
This is harder to spot than it sounds, because the most recent AI image generators have largely fixed the obvious errors that made earlier fakes detectable. The distorted hands, the extra fingers, the melting ears, the teeth that looked like a single smooth tile — many of these artifacts are gone in the systems being used in 2026.
What remains, and what trained reviewers still notice, tends to be subtle. The face in an AI-generated photo often has an uncanny smoothness, a kind of statistical averageness where every feature is conventionally attractive but the overall effect is slightly frictionless. Background elements sometimes warp near the edges of the subject. Reflections in glasses or eyes may not be consistent with the stated environment. Hair, especially around the temples and in motion, is still an area where AI generators occasionally produce results that look processed rather than natural.
Norton reported that only 46% of people correctly identified AI-generated photos in a test, which means the majority of people looking at these images in a normal dating context, without any reason to be suspicious, will see something that looks real.
A single AI-generated photo is suspicious because real people have multiple photos. Scammers know this, so organized operations now generate sets of five to ten images of the same synthetic face in different settings: outdoors, at a cafe, with friends (other AI faces), on a weekend trip. The facial consistency across generated images has improved significantly, though it remains an area where careful comparison sometimes reveals subtle mismatches in facial geometry between shots.
Scammers use video calls strategically, not always avoiding them. The approach depends on what tools the operation has available.
The three main methods are pre-recorded video loops played through fake webcam apps, deepfake AI technology that animates a face in real time, and brief calls that end before the victim can request verification.
The pre-recorded loop method is the oldest and still common. The scammer plays footage of a real woman (recorded without her knowledge or obtained through other means) through software that injects it as a fake camera feed into WhatsApp, Telegram, or standard video call platforms. The woman appears to be live because she is moving and speaking, but the feed is a loop or a recorded clip playing on repeat.
Real-time deepfake technology, which was largely theoretical for consumer-grade fraud two years ago, is now accessible enough to be used in documented cases. The FBI Norfolk Field Office confirmed in February 2026 that criminals now use AI-generated voice messages to make schemes more believable, and that scammers can appear on a live video call using a completely fabricated face. The visual quality of these real-time deepfakes varies, but because the call happens in a context of established trust and emotional investment, victims are not looking for flaws. They are looking for the person they believe they know.
This is the most important practical point in this article. If someone agrees to a video call, it does not confirm they are who they say they are. That assumption was reasonable until recently. It is not reasonable now.
A video call is still useful as a verification tool, but only if you use it actively rather than passively. Watching someone talk does not tell you anything definitive. Asking them to do something specific and unpredictable, on the other hand, still reveals a great deal.
|
Behavior |
What It Looks Like |
Why It Matters |
| Photos too consistent | Every photo is well-lit, well-composed, same flattering angle | Real people have bad photos. AI generators do not. |
| No social media footprint | Username search returns nothing; no Facebook, VK, Instagram for a person in their 30s | Real people leave digital traces. Synthetic profiles are built specifically for one platform. |
| Video call quality always degrades at key moments | Pixelation or freezing whenever you make a specific request | Consistent with a pre-recorded loop struggling with an unexpected input |
| Cannot perform an unscripted action on video | You ask her to wave with her left hand or hold up today’s newspaper and the call drops | Pre-recorded content cannot respond to real-time instructions |
| Moves off the dating platform within days | Pushes to WhatsApp or Telegram before the first week is over | Platform moderation cannot follow. The account can be deleted and the conversation disappears. |
| Profile was recently created | Account is less than a few weeks old and has few connections or history | Consistent with a freshly built fake Russian dating profile |
| Messages arrive at statistically consistent intervals | Responses come within a narrow time window regardless of time zone or day of week | Consistent with scripted or AI-assisted message management |
Run the photo through all three image search engines. Google Images, Yandex Image Search, and TinEye each index different content. Yandex is particularly effective for Eastern European sources. A result does not guarantee authenticity, but the absence of results no longer guarantees a real person either, because AI-generated faces have no source to find.
Use an AI image detection tool on the profile photo. Several free tools analyze images for signs of GAN generation or diffusion model creation, including anomalies in skin texture, background coherence, and metadata. These tools are not infallible, but they add a meaningful layer of detection that a simple eye test cannot provide.
Request a spontaneous, specific action on video. Ask her to hold up three fingers and say today’s date out loud. Ask her to write your name on a piece of paper and show it to the camera. Ask her to stand up and turn around slowly. Current real-time deepfake algorithms still struggle with complex hand gestures and edge blending, leading to visible errors or flickering when the subject is asked to perform an unscripted action. A genuine person can do any of these things in under thirty seconds. A pre-recorded loop or a poorly maintained deepfake cannot.
Check the Russian Scammers Blacklist at russian-women-blacklist.com for reported usernames, phone numbers, and profile photos from documented cases. A match is not necessary to indicate a scam, but a match is a clear signal to stop.
Verify documents independently, not emotionally. If a passport image is sent to prove identity, do not treat it as reassurance. Passport images are trivially easy to generate or forge and are used in scams precisely because they feel official.
Slow down without signaling suspicion. Tell her you have been busy and will be less available for a while. Observe whether the pressure increases when you become less responsive. Scam operations running multiple targets simultaneously often apply more intensity when a target goes quiet, because the operation needs to close before emotional investment fades.
Request the spontaneous video test described above. Frame it as wanting to “feel more connected” rather than as a verification attempt. A genuine person will find this easy and maybe slightly amusing. A scammer using pre-recorded content will find an excuse.
If you have suspicions you want to discuss without alerting your contact, the community forum at forum.allaboutdatingscams.com has documented hundreds of Russian online dating scam patterns and can help you compare what you are seeing against known cases.
Do not send money for any purpose before meeting in person. Travel expenses, visa fees, medical emergencies, legal problems, account freezes: these are the standard catalogue of pretexts used across documented cases. The specific story varies. The financial request does not.
Contact your bank immediately and report the transfer as fraudulent. Wire transfers and cryptocurrency transactions are difficult to reverse but not always impossible if you act within the first 24 to 48 hours. The FBI’s Financial Fraud Kill Chain has successfully frozen funds in transit in documented cases.
Report the case to the FBI’s Internet Crime Complaint Center at ic3.gov. Detailed reports, including usernames, phone numbers, email addresses, the platforms used, and the full message history if you have it, help the FBI connect individual cases to larger operations and take action. You can also report to the FTC at reportfraud.ftc.gov.
Do not pay any further money to recover what you have already lost. A secondary scam commonly follows known victims: a supposed recovery agent or legal representative contacts them offering to retrieve the funds for an upfront fee. This is a separate fraud aimed at the same person.
The shift from stolen photos to AI-generated photos is not a minor update to an old scam. It removes the most reliable detection method most people had. Americans lost an estimated $3 billion to romance scams in 2025, up from $1.2 billion in 2024, and AI-enabled fake Russian dating profiles are a significant contributor to that figure. The response is not to stop using dating platforms. It is to change how you verify the person behind the profile before any emotional or financial investment is made. A real person can hold up a piece of paper with your name on it. An AI-generated persona, however sophisticated, cannot.
