Safety
How to tell if a dating photo is AI-generated or fake (and why an empty search proves nothing)

No single check proves a dating profile photo is real, and a search that finds nothing proves the least of all. In our November 19, 2026 test, a mirror, crop and slight brighten hid 13 of 15 photos from TinEye, even though TinEye had found every one of the unedited originals. A face generated by AI has no original to find in the first place. So the reliable method is a sequence: search with two engines, flip the photo and search again, study the face for tells, and then ask for something live.
Why “I searched and found nothing” isn’t an answer
Most advice on this topic stops at “run a reverse image search.” The FTC gives the same advice in its romance scam guidance: do a reverse image search of the person’s profile picture, and treat a different name or mismatched details as a sign of a scam. That’s good advice for the most common fake, a stolen photo set. But it quietly teaches people to read a blank result as a green light.
Our thesis is that a blank result is just a blank. It can mean the photos are private. It can mean the photo was edited enough to slip past the engine. And it can mean no camera ever took the picture. We tested the middle case directly. We took 15 portraits that we knew were findable, made one light edit to each, and searched again. Most of them vanished from the engine built specifically to find edited copies. If you’re new to the search itself, our reverse image search guide has the step-by-step. This page covers what to do when that search comes back empty.
The check sequence, in order
Run these in order. Each step catches something the one before it can miss, and none of them is conclusive alone.
- Reverse search with two engines. Use Google Lens first and TinEye second, on the full photo and then on a tight crop of the face. Two engines matter because they miss different things (see the table below). The exact steps are in the method on our reverse image page.
- Flip the photo and search again. Mirror the image horizontally, crop in a little, and run both engines once more. This is the edit a scammer would make, and it’s the one that defeated TinEye on 13 of our 15 photos. Most phones can mirror a photo in the editing screen.
- Compare every photo on the profile. Does the face stay the same shape across angles, lighting and years? Do all the pictures share one polished look, or one lens? Real people have awkward shots. A set of near-identical, glossy portraits deserves a harder look.
- Look for the visual tells. Zoom in on hands, teeth, earrings, glasses, hair edges, background text and lighting. The list is below, along with a warning about how far to trust it.
- Check them live. Ask for a short video call, or for a photo that can’t be prepared ahead of time. This is the step a fake can’t easily pass, and it’s the only one that tests the person rather than the picture.
The visual tells of an AI-generated face
- Hands. Extra, missing, merged or oddly bent fingers, or a hand that’s hidden in every photo.
- Teeth. A row that’s too even, blurred together, or a smile that shows more teeth than a mouth should hold.
- Earrings, glasses and necklaces. A pair that doesn’t match, an earring that fuses into the ear, or glasses frames that change shape across the face.
- Hair. Strands that melt into the skin or the background, or an edge that looks painted.
- Background text and objects. Signs and labels with garbled letters, warped door frames, furniture that doesn’t quite resolve into a real object.
- Lighting. Shadows on the face that fall one way while the scene’s light comes from another, or skin that’s smooth in a way real pores aren’t.
Treat these as weak evidence in one direction only. A visible tell is worth noticing. The absence of tells proves nothing, because current image generators fix many of them, and an ordinary retouching app can produce some of the same smoothness on a real face. Free AI-image detectors can add a clue, but don’t treat a percentage as a verdict, since cropping and compression change what they report.
The live check
Ask for a short video call before you share a phone number or plan a meeting. If a call isn’t possible yet, ask for a quick photo that can’t be made in advance: the person holding up a specific number of fingers, or a piece of paper with today’s date and your first name. Then compare the face to the profile photos. A real person can usually do either in a minute.
One caution so you don’t overread it. A refusal isn’t proof of fraud, because plenty of honest people dislike video calls. A pattern of excuses across several asks is a different thing, and it’s a documented feature of romance scams. If a call does happen, ask them to turn their head side to side or pass a hand in front of their face. Live face filters can break under that kind of movement, though we haven’t tested it ourselves, so treat it as a bonus and not a guarantee.
What our test found
In NexSpark’s November 19, 2026 test of 15 Unsplash portraits, Google Lens found 15 of 15 originals but only 9 of 15 mirrored, cropped and brightened copies. TinEye found 15 of 15 originals and just 2 of 15 edits. Bing found 3 and 1. Yandex blocked us with a CAPTCHA, so it has no results here.
| Search engine | Originals found (of 15) | Edited copies found (of 15) | What it tells you |
|---|---|---|---|
| Google Lens | 15 | 9 | Best single engine. Five of the six misses locked onto clothing or jewelry, not the face. The sixth returned only tag pages. |
| TinEye | 15 | 2 | Perfect on untouched photos, nearly blind to a mirror, crop and slight brighten. 13 edits returned zero matches. |
| Bing Visual Search | 3 | 1 | Mostly lookalike photos and generic pages. Not a reliable second opinion. |
| Yandex | Not tested | Not tested | Stopped by a bot check on the first search. We didn’t try to get around it. |
Counting all three engines together, 10 of the 15 edited photos turned up somewhere, which means 5 slipped past every engine we could use. That’s our own tally from the test data, and it’s the number to keep in mind: one light edit was enough to leave a third of known, findable photos with no source at all. We scored a hit when the engine returned the Unsplash photo page or another page showing the same photo; tag or listing pages counted only when the photo itself appeared on them. We rescored two results on November 19, 2026 to apply this rule strictly: Bing on the original of photo 13 and Lens on the edited copy of photo 14 had returned only tag pages, so they now count as misses. The full method is on the reverse image search page.
Limits of this test. It used 15 stock portraits on one day, one edit recipe, and a snapshot of engines that change constantly. We did not test AI-generated faces. That part of the thesis is reasoning, not a measurement: an image that was never published can’t have an earlier copy to match. We also didn’t test paid face-search tools.
A real photo isn’t a real profile
Passing every photo check still doesn’t clear the profile. A stolen set of real photos will look perfectly natural, because it is. If a search turns up the same face under another name, a stock site or a model’s portfolio, you have your answer. If it turns up nothing, you’ve learned only that you haven’t found the source yet. For what happens after the photos pass and the story still shifts, see our catfishing signs guide.
What ID verification adds
Everything above is you doing a verifier’s job by hand, one profile at a time. NexSpark verifies every member’s ID with a government photo ID and a selfie before they’re introduced to anyone, which makes a borrowed or generated face far harder to pass off. You can read how verification works for the details.
If you think the photo is fake
Stop sharing personal details and don’t send money. The FTC’s bottom line is to never send money or gifts to a sweetheart you haven’t met in person. Screenshot the profile and the conversation, report the account in the app, and if money has already moved, report it at reportfraud.ftc.gov and tell your bank right away.
Frequently asked questions
How can you tell if a dating profile picture is AI-generated?
Zoom in on hands, teeth, earrings, glasses, hair edges, background text and lighting, and compare the face across every photo on the profile. Visible flaws are a clue, but current generators fix many of them, so a clean-looking photo proves nothing. Back the visual check with a reverse image search on two engines and a live check, such as a short video call.
Is a dating profile photo real if a reverse image search finds nothing?
Not necessarily. An empty result can mean the photos are private, that the picture was edited enough to slip past the engine, or that it never existed before because an AI made it. In NexSpark’s November 19, 2026 test, a mirror, crop and slight brighten hid 13 of 15 known photos from TinEye, so a blank result only means you haven’t found a source yet.
Why does flipping a photo help a reverse image search?
A mirrored and slightly cropped copy is the edit a scammer would make, and it can defeat engines that match the original. In our test of 15 portraits, TinEye found all 15 originals but only 2 of the 15 edited copies, while Google Lens found 9. Flip the photo, crop in a little, and search again on both.
Which reverse image search is best for dating profile photos?
In our test, Google Lens was the strongest single engine: 15 of 15 originals and 9 of 15 edited copies. TinEye matched all 15 originals but only 2 edits, and Bing found 3 and 1. Yandex blocked the test with a bot check. Use Google Lens first and TinEye second, and don’t treat either as proof.
Can an AI image detector tell me if a dating photo is fake?
It can add a clue, but it shouldn’t decide the question. Cropping, filters and compression change what these tools report, and a real photo can still belong to a fake profile if someone stole it. Use a detector’s score as one input next to a reverse image search and a live check.
What’s the safest way to confirm someone is real before meeting?
Ask for a short video call, or for a photo that can’t be prepared ahead of time, such as holding up a specific number of fingers or a note with today’s date. A pattern of excuses across several requests is a warning sign. The FTC’s bottom line is to never send money or gifts to someone you haven’t met in person.
Sources
- NexSpark reverse image search test, November 19, 2026: 15 Unsplash portraits, originals and edited copies, Google Lens, TinEye, Bing Visual Search and Yandex. Method and notes on our reverse image search page.
- Federal Trade Commission, What To Know About Romance Scams: reverse image search of profile pictures; never send money or gifts to someone you haven’t met in person.
- Federal Trade Commission, ReportFraud.ftc.gov, for reporting a suspected scam.
Not sure the person you’re talking to is real?
Four questions on what you can verify, whether they’ll appear on video, who pushed to move off the app, and whether money has come up. No signup required.
Skip the profile detective work
Matched on values and compatibility, with every introduction reviewed by a real matchmaker. Start with a few honest questions, and your ideal match report is waiting at the end.
By