A message arrives from a recruiter you’ve never heard of. The profile photo is polished — studio lighting, perfect skin, a confident smile that could belong to any senior director. The company looks legitimate. The opportunity sounds real. What you may not realise is that the face looking back at you was never a person.
AI-generated headshots are now a standard tool in the playbook of fake recruiter scams, sales bot campaigns, and identity theft operations on LinkedIn. The platforms are slow to catch them and the visual quality has crossed the threshold where most people won’t look twice. This article explains what to look for, why the most common counter-check fails completely, and how to get a forensic answer in seconds.
Why fake LinkedIn profiles use AI headshots
Creating a convincing fake identity used to require stealing a real person’s photo — which was detectable through reverse image search. AI-generated faces solve that problem entirely. The attacker gets a portrait of a plausible-looking professional that has never existed, which means it has no prior online presence to betray it.
The use cases are varied and growing. Fake recruiters lure candidates into phishing flows or harvest resumes for credential stuffing. Sales bots impersonate business development professionals to extract company information. State-level actors use fabricated identities to build influence networks inside organisations. In every case, the AI headshot is the first credibility signal the target evaluates.
The GAN tell: eyes locked to centre
The most consistent artifact of GAN-generated faces — the architecture behind tools like StyleGAN, which powers sites such as thispersondoesnotexist.com — is that the eyes tend to appear at nearly the same position in the frame, almost always near the vertical centre of the image.
This is not coincidence. GANs are trained on large face datasets where the eyes are roughly aligned to anchor the latent space. The result is a subtle tell: look at a gallery of AI headshots and the pupils cluster at a predictable height. Real portrait photographers vary framing constantly.
Look also at eye-level symmetry within a single image. Generated faces often show:
- Pupils that aren’t quite round, or that differ slightly in size between left and right
- Catchlights (the small white reflections in the iris) that appear in one eye but not the other, or that reflect a light source inconsistent with the rest of the scene
- A glassy, slightly unfocused quality to the gaze that doesn’t resolve even when you zoom in
Background blur artifacts and boundary breakdown
A phone camera creates smooth, optically consistent bokeh. A GAN-generated image creates a learned approximation of bokeh, and it breaks down at edges. Inspect the region where the subject’s hair, ears, and shoulders meet the background. Signs to look for:
- Hair strands that blur into the background rather than maintaining individual threads at the boundary
- An unnaturally smooth gradient between the subject and background, as if the two were composited rather than photographed together
- Repeated textures in the background — wall patterns, foliage, office furniture — that tile or smear in ways a real photograph would not
Asymmetric accessories: the single earring problem
GANs struggle with asymmetric objects. A face is roughly symmetric and the model handles it well; accessories that should appear on only one side introduce an asymmetry the model must learn rather than reflect. The result is a repeatable class of errors:
- Earrings that appear on one ear but are missing, distorted, or morphed into the hair on the other
- Glasses frames where one lens is rendered cleanly and the other shows warping, extra thickness, or blends into the skin at the temple
- Collars and lapels that maintain consistent geometry on one side but ripple or blur on the other
This is one of the more actionable visual checks because it doesn’t require zooming into fine texture — the asymmetry is often visible at thumbnail scale.
The single-photo account: a compound risk signal
Most of the forensic signals above require you to examine the photo closely. There is a higher-level signal that requires no image analysis at all: the account has exactly one photo.
Real professionals on LinkedIn accumulate a visual history. A conference panel shot from three years ago. A team photo from an offsite. A headshot that replaced an older headshot. The absence of this visual archaeology — a brand-new account with a single, impeccably lit portrait and nothing else — is a compound risk signal.
It doesn’t mean the person is fake. It means the cost of scrutiny is low and the potential consequence of not scrutinising is high.
Why reverse image search gives false safety
The instinct to run a reverse image search is reasonable and the conclusion people draw from it is almost always wrong.
Reverse search works by finding exact or near-exact pixel matches in indexed web content. A stolen photograph of a real person will often surface — which is a genuine catch. A freshly synthesised AI face has never appeared anywhere online, so the search finds nothing, and “found nothing” gets read as “this person is real.”
This is not a criticism of reverse search — it catches a real class of fraud. The problem is using it to answer a question it was never designed to address.
How FakeRadar approaches LinkedIn headshot analysis
When you upload a headshot to FakeRadar, the analysis runs across several independent layers rather than issuing a single verdict:
| Layer | What it checks | What a GAN headshot tends to show |
|---|---|---|
| Face-swap detector | Face boundary integrity and facial geometry | Boundary artifacts, geometry inconsistencies |
| AI model score | Learned generation patterns | Elevated generation probability |
| Frequency (FFT) | Pixel-level spectral fingerprint | Periodic grid artifact from GAN upsampling |
| Error Level Analysis | Compression uniformity across regions | Uniform compression that real camera photos don’t produce |
| Metadata / EXIF | Camera origin signals | Missing or generic EXIF, no device fingerprint |
The result is reported as signals detected or not detected — not as “this is fake.” That distinction matters because no detection layer is infallible, and because wrongly accusing a real person of using a fabricated photo is a serious outcome worth avoiding.
If you want to understand the broader principle behind this approach, the article on why AI detection is signal-based covers it in depth.
A responsible frame: scrutiny without accusation
It is worth being direct about what this kind of analysis can and cannot do. A high AI-signal score on a LinkedIn headshot does not mean the person is running a scam. Many legitimate professionals now use AI-assisted headshot tools (PhotoAI, Aragon, and similar services) to generate polished portraits — the same underlying technology, used transparently for personal branding.
The right frame is a photo authenticity check, not a fraud accusation. The question to answer is: does this image carry the forensic signatures of a synthesised face? If the answer is yes, that warrants more scrutiny of the full profile — cross-referencing the claimed employer, checking whether the posting history is consistent with a real professional’s timeline, and potentially verifying identity through a video call before sharing sensitive information.
The face-swap detection guide covers the broader detection landscape, and how to tell if a dating profile photo is AI applies many of the same methods to another high-risk context.
FAQ
Can you tell if a LinkedIn profile photo is AI-generated just by looking?
Sometimes, but not reliably. The tells are subtle at thumbnail scale — eye symmetry, background blur at hair edges, asymmetric accessories. Forensic analysis is the only method that goes below the visual surface.
Does reverse image search catch fake AI LinkedIn photos?
No. A newly generated AI face has never existed online, so reverse search returns nothing. That absence is misread as proof of authenticity. Reverse search catches stolen real photos; it is blind to synthesised ones.
What are the most common visual signs of a GAN-generated headshot?
Eyes at nearly identical frame height, background that smears near the shoulders and ears, accessories that are asymmetric or partially rendered, and skin so smooth it lacks pores or fine texture.
Is it safe to connect with a LinkedIn profile that has only one photo?
One photo is a risk signal, not a verdict. Real professionals accumulate visual history across years. A new account with a single perfect headshot and no other imagery warrants additional verification before sharing information or engaging with any requests.
Summary
- GAN-generated headshots have repeatable artifacts: centred eye positioning, background blur at subject edges, asymmetric accessories, and overly uniform skin.
- Reverse image search gives false safety — it finds nothing for synthetic faces because there is nothing to match.
- A single photo on a new account is a compound risk signal worth further scrutiny.
- Forensic analysis stacks independent layers — face-swap detection, AI model score, frequency analysis, ELA, and metadata — and reports signals, not verdicts.
- The right frame is a photo authenticity check; wrongly accusing a real person of using a fake photo is a serious outcome to avoid.
- Run a forensic check when the visual pass leaves you uncertain or the stakes of being wrong are high.
Try it yourself
Analyse an image or video free
Found a suspicious LinkedIn photo? Upload it and FakeRadar runs a forensic face-swap and AI-generation check — no account needed.
Start Free Analysis