How to tell if a dating profile photo is AI-generated (and why you often can't)

By the Authentk team  ·  Published  ·  Updated

The profile looks fine. Good lighting, a real smile, a wedding photo and one with a dog. The conversation is easy. And somewhere in the back of your mind is the question everyone on a dating app asks now: is this a person, or a face a computer made? The advice you will find online says to look at the hands, the earrings, the text on a shirt, the background. Most of that advice is out of date. This article covers what the research says about spotting AI-generated faces, which checks still work, what the apps' verification badges actually prove, and how to ask someone for proof without sounding paranoid.

Why the visual tells stopped working

For a few years, generated faces had obvious problems. Earrings that did not match. Backgrounds that melted into hair. Teeth that ran together. Hands with the wrong number of fingers. Those tells were real, and they are mostly gone. The tools improved, and a scammer needs only one clean image, picked from hundreds, to open a profile. The stakes are not small: in 2023, Americans reported more than $1.1 billion in losses to romance scams, with a median loss of $2,000 (FTC), and 2 in 5 current U.S. dating-app users say they have been targeted by a dating scam (Norton, 2025).

The research on how well people spot these faces is blunt. In a 2022 study published in PNAS, Sophie Nightingale and Hany Farid showed 315 people a mix of real photos and faces generated by StyleGAN2 and asked which were which. Average accuracy was 48.2%, a coin flip. A second group was trained on what to look for and given feedback after every answer, and still reached only 59.0%. In a third experiment, the generated faces were rated 7.7% more trustworthy than the real ones (Nightingale and Farid, PNAS, 2022).

A 2023 paper in Psychological Science found something stranger. White AI-generated faces were judged to be human more often than photos of real people, an effect the authors call AI hyperrealism and attribute to the faces the models were trained on. The people who were worst at the task were also the most confident. A machine classifier in the same study reached 94% on the same faces, which tells you where this is heading: a job for software, not eyes (Miller et al., Psychological Science, 2023).

The practical conclusion: if a profile photo looks slightly off, trust that feeling. If it looks fine, that tells you nothing at all.

Three AI-generated portrait photos of people who do not exist: a smiling woman with shoulder-length hair, a man with short dark hair, and a young woman with a fringe

Three faces generated by StyleGAN. None of these people exist. Images: Wikimedia Commons, public domain (AI-generated works).

What you can actually do today

Five checks, in rough order of effort. The limits matter as much as the checks.

1. Reverse image search. Google Lens, Bing and TinEye will show you where else a photo appears online. This still catches the older, still common scam: real photos lifted from a model, a soldier or an influencer and reused under a new name. The limit: a generated face has never been published anywhere. A clean result tells you the photo is not stolen. It does not tell you the person exists.

2. AI image detectors. Several websites will take an upload and return a score. Treat the score as a weak hint, not a verdict. Detectors are trained against particular generators and lag the newest ones, and the copy you have from a dating app has been cropped, compressed and resized, which strips much of what the detector was looking at. If a detector says 90% likely AI, be more careful. If it says 10%, do not relax.

3. The app's verification badge. The big apps all run a selfie check now, and they are worth using. Tinder's Photo Verification has required a video selfie, "a series of video prompts," since April 2023, replacing still-photo poses (Match Group). Bumble asks the profile owner for "a real-time selfie in a randomly-chosen pose," which a reviewer compares with the profile photos, and you can ask a match to verify from the chat screen (Bumble). Hinge's Selfie Verification is optional: a video selfie that machine learning and staff compare with the profile photos (Global Dating Insights, 2022). What the badge proves: the person who ran the check matched the photos on the profile on the day they verified. What it does not prove: anything about a photo they send you next week, or who is typing right now. Match Group's own wording is "while no photo verification process is perfect." Use the badge as a filter, not a guarantee.

4. A video call. Still the strongest everyday check. Ask for ten minutes on camera and ask them to turn their head, wave, or pick up a mug. A still image cannot do that. The limit: real-time face-swap tools exist and are improving, so watch for lag, edges that blur when they turn, and a reluctance to move. The scammer's standard replies are a broken camera, a bad connection, a posting overseas. Each of those is an answer in itself.

5. A gesture selfie. Ask for a selfie right now doing something you choose, not something from a list: three fingers up, holding a spoon, a thumbs-down. A generated face cannot comply quickly, and an old photo cannot match. The limits: a determined scammer with editing tools has a few minutes to work with, a photo of a real person does not prove they are the one you will meet, and you have no way to see from the file whether it was taken just now or last year. More than half of young daters have met someone who looked different from their photos (Hily, 2025). A gesture selfie narrows that gap. It does not close it.

Proof at capture: what a signed selfie adds

Every check above is a judgement after the fact. You look at a photo and try to decide whether it is real. The alternative is to fix the moment the photo is taken, so there is nothing left to judge. That is the idea behind a cryptographically signed selfie, and it is what we built Authentk to do. The other person opens the app and takes one selfie. Their phone signs it against a real-time challenge, so the photo can only have been taken at that moment, on that device. What you receive is a watermarked photo with the capture time and a verification link. Open the link or scan the QR code to see when it was taken, then upload the photo they sent you. You see "Photo Matches" if it is the signed original, or "Photo Does Not Match" if it was edited or swapped. You do not need the app to check, and checking costs you nothing. See what a verified photo looks like.

The limits, stated plainly. A signed selfie proves the photo is real, was taken just now and came from that person's device. It does not prove a name, an age or an intention, and it is not an ID check. It needs the other person to have an iPhone and to be willing, which most honest people are. It is one option among the five above, not a replacement, and the right moment for it is the same as for a video call: before you meet.

Authentk proves a selfie is real: taken now, by that person, unedited. Free iPhone beta.

Join the iPhone beta

How to ask without sounding paranoid

Most people skip the good checks because asking feels like an accusation. It is not, if you say it the same way every time and offer to go first. Copy these.

Honest people grumble and do it. The replies that should end the conversation are the broken camera, the deployment, "after we meet," and any version of "don't you trust me." "Not yet" is a fine answer to that. The same routine works anywhere a stranger's photo is all you have to go on. Related: Facebook Marketplace meetup safety: 9 rules before you meet and How to verify a roommate is real before paying a deposit.

Questions

Can you tell an AI-generated dating photo just by looking?

Usually not. In a 2022 study, people sorted real faces from generated ones with 48.2% accuracy, about a coin flip, and training only raised that to 59%. A photo that looks slightly off is worth trusting. A photo that looks fine tells you nothing.

Does a reverse image search catch AI-generated profile photos?

No. Reverse image search finds photos that already exist online, which is useful for stolen pictures of real people. A generated face has never been published anywhere, so a clean result is not evidence that the photo is genuine.

Does a Tinder, Bumble or Hinge verification badge mean the photos are real?

It means the person who ran the check matched their profile photos at the time, usually through a video selfie or a posed selfie reviewed by staff. It does not cover photos they send you later, and Match Group itself says no photo verification process is perfect.

What is the most reliable way to check that a dating match is real?

A short video call where you ask them to move, or a selfie taken right now doing a gesture you choose. A cryptographically signed selfie goes further: it proves the photo was taken at that moment on that person's device and was not edited, and anyone can check it from a link without an app.

The short version

Stop hunting for extra fingers. People sort real faces from generated ones at about a coin flip, and the generated ones look more trustworthy. Use reverse image search for stolen photos and the app's badge as a filter, then ask for the one thing a generated face cannot give you: something that happens right now, on your terms. A video call with movement, a gesture you chose, or a selfie signed at the moment it was taken.

Sources

© 2026 Authentk. All rights reserved.  ·  Pat. Pend.  ·  Home  ·  Blog  ·  About  ·  Privacy Policy  ·  Terms of Service