Is this profile photo AI? How to check a photo of a person
The short answer
A photo of a person can be AI-made, or it can be a real person's photo that someone else is using. A check of the picture file can find the first (if the file says so, or a detector flags it) but not the second, and the second is the more common way fake profiles are made. So check the file, search for the same photo elsewhere with a reverse image search, and look at the account around it. A tool cannot tell you who is in a photo, and a score is not a reason to accuse anyone.
You get a friend request, a message from a stranger, a job applicant's headshot or a seller's profile picture, and something feels off. The question people type is "is this person AI?", but it hides two questions. Was the picture made by a generator? And is the person who they say they are? A picture check can only speak to the first, and often not even that. This guide gives a routine that covers both and says where each step stops.
Two ways a profile photo can be fake
| AI-generated face | Stolen real photo | |
|---|---|---|
| What it is | A face that never existed, made by an image generator. | A real person's photo, taken from their account or a website and used by someone else. |
| Can a file check find it? | Sometimes: if the file says so, or a detector flags it. | No. It is a real photo, and every signal says so. |
| What finds it | The checks below, plus context. | A reverse image search, and the account's history. |
Both are in use. Do not assume that a picture that "passes" is safe: a stolen photo passes every AI check, because it is genuine.
Step 1: check the file
Get the original file if you can: a screenshot or a copy that went through a messenger has lost its metadata. Then open it in our checker for photos of people (it is the same engine as the AI Image Checker). Read the reasons and not only the headline:
- "AI-made: the file says so" means the file names a generator, or carries a prompt, a workflow, Content Credentials or an AI label. That is solid for that file.
- "Looks like a real camera photo" means the file carries camera records: make, model, lens, settings, the exact time. It is good evidence the picture started as a photograph, but the records can be copied and a real photo can be edited afterward.
- A score is a guess from how the picture looks. In our test a detector flagged 85% of AI pictures and 25% of real ones, and the two bands in the middle were wrong for one picture in five or more. How we measured it, and the numbers for each band, are in the guide to telling if an image is AI-generated.
Why faces are the hardest case
Faces get the best-trained generators, and the old giveaways have been fading: mismatched earrings, glasses that melt into the skin, hair that merges with the background, garbled text on a shirt, a hand with the wrong number of fingers. Older generators did these things, and some pictures still do, so they are worth a glance. But newer models make fewer of them, and a good portrait can be photographic in every detail. Looking harder is not a method; checking the file and the context is.
Our own test shows the limit. We had only five real photos of people and five faces from an early face generator, too few for a reliable rate, but enough to say: treat a score on a face as less certain than a score on a landscape, and do not read a middle score as a lean in either direction.
Step 2: search for the same photo elsewhere
This is the step that catches stolen photos, and it works on any picture. Search for the image on TinEye or with Google Lens. If the same face appears on a different name, a stock photo site or a photographer's page, the profile is using someone else's picture. If nothing turns up, that proves little: a new photo, or a private one, will not be in the index, and a generated face never will be. Note that uploading a picture to those sites sends it to them. Our checker is a separate step, and the copy it sends to us is described in our privacy policy.
Step 3: read the account around the picture
The photo is one clue. The account is another, and it is usually more telling:
- Age and activity. A profile created last week with a polished picture, few posts and many requests is worth caution.
- Consistency. Do the name, the job, the place and the other photos fit together? A set of different pictures of the same face is harder to fake than one portrait, though not impossible.
- The ask. Pressure to move to another app, to send money or to share a code matters more than the face. Those are the habits of scams, whoever is behind the picture.
- A live check. Ask for a short video call, or for a new photo of them holding something you name. A person can do that in a minute and a generated profile usually cannot.
What not to conclude
- Not "this person is not real". A real person can have an AI-looking photo (a retouched or stylized one), and a flat score is not an identification. We do not identify anyone and our tool does not compare faces.
- Not "this person lied". Plenty of people use an AI-made or edited picture for privacy or fun. It is a reason to be careful about trust, not an accusation.
- Not "it passed, so it is safe". A genuine photo can belong to the wrong person.
If you must report a profile, report what you observed: "the photo appears on another site under a different name", or "the file carries a prompt from a generator". Those are facts a platform can check. "A detector said 90" is not.
Find out whether a picture was made by AI: it reads the prompt, workflow, Content Credentials and invisible watermark in the file, looks for camera records, and estimates how much it looks AI-made. It runs in your browser. Open the AI Image Checker.
Frequently asked questions
Can an AI checker tell me who is in a photo?
No, and ours does not try. It looks at the picture file and its pixels. It does not recognize or compare faces, and a result says something about the file, never about a person.
What if a real photo is flagged as AI?
It happens: a quarter of the real pictures in our test were flagged by the detector, anime art and paintings most often. Look at the reasons, check whether the file carries camera records, and do not treat a score as proof of anything.
Do most fake profiles use AI faces?
We have no measurement of that. Both AI faces and stolen real photos are used, and a stolen photo passes every AI check, which is why a reverse image search and the account's history matter as much as the picture.
Is a photo with no metadata suspicious?
No. Messengers, social sites and screenshots remove metadata from real photos all the time. A missing record says nothing either way. A present one can say something: camera records point to a photograph, and a generator's prompt or Content Credentials point the other way.