Can SynthID detect your AI LinkedIn photo?

What did Google's new SynthID site actually launch?

Google opened its SynthID Detector to the public on 6 October 2026, so anyone can now upload an image, video or audio file and ask whether it carries Google's invisible AI watermark. Until then the portal was limited to a small group of journalists, researchers and media professionals. It's free, available worldwide in English, and you sign in with a Google, OpenAI or Apple account. The image side accepts JPG, PNG, WEBP, BMP, AVIF, HEIC, TIFF and GIF; video covers MP4, MOV and WEBM; audio takes WAV, MP3, OGG, FLAC, AAC and M4A. Results aren't a bare yes or no, either — the portal can highlight which regions of an image or which segments of a clip look watermarked. Google says people already make around a million verification requests a day.

The watermark only exists in files produced by models that chose to embed it. That means Google's own stack — its image, video and music generators, plus Gemini and its creative tools — and a growing partner list that includes OpenAI, NVIDIA and Kakao, with Apple reported to be joining. OpenAI runs its own checking page, and Google has wired verification into the Gemini app and Chrome. Read the verdict wording closely, though: a positive result says the media was made or edited with Google AI, which lumps a fully synthetic face together with a photo you merely cleaned up in an AI editor. The portal also warns that the file may have changed since the watermark went in.

What it can't do matters more for your job search. Midjourney, Stable Diffusion and most dedicated headshot apps running their own fine-tuned models carry no SynthID mark, so a clean result proves nothing at all — the absence of a watermark is not evidence of a camera. Heavy crops, screenshots, format conversion and aggressive re-compression can strip or weaken the signal. And while SynthID does watermark Gemini's text output, the public portal is built around images, video and audio, so pasting your CV into it simply isn't an option. No detector on the market is reading your bullet points today and returning a verdict you should trust.

Can a recruiter tell your AI headshot isn't real?

Most can't — not reliably, and not from a thumbnail. In the largest independent test of this, 1,087 recruiters were asked to pick out the AI-generated headshot and got it right only 39.5% of the time, even though 80% of them had said beforehand that they could spot a fake. Against premium generators their accuracy dropped to 29.2%; against free, low-quality tools it climbed to 58.9%. In blind comparisons, 76.5% actually preferred the AI headshot, rating it more professional and more approachable than the real photograph. That survey ran in 2024, before the current generation of image models, so assume the human eye has lost ground since, not gained it.

Here's the part the headshot vendors quote rather less often. In the same study, 66% of recruiters said they'd be put off a candidate once they learned the photo wasn't real, and 88% thought candidates should disclose it. The odds are stacked in a specific, uncomfortable way: you probably win the three-second glance and lose the relationship the moment it comes up. The tells people do catch cluster around secondary detail — warped backgrounds and jewellery, plastic-smooth skin, uncanny facial symmetry, a blur that's uniform everywhere instead of falling off with depth. At 200 pixels wide, none of that is visible. On a 1080p video call, all of it is.

My honest take: the detector isn't your real risk, and neither is a recruiter squinting at your thumbnail. The first video interview is. A face that doesn't match the profile reads as deception even when you never intended it, and that's a far more expensive outcome than a slightly dated photo. So retouch your real face as much as you like — better lighting, a cleaner background, a shirt that isn't creased. Don't commission a new one. This matters more in Israel than people assume: Israeli CVs usually go out without a photo, so your LinkedIn picture carries the whole visual impression a recruiter forms before the call, and it's the one they screenshot into a shortlist.

Does LinkedIn allow an AI-generated profile photo?

Yes, provided the person in the picture is recognisably you. LinkedIn's profile photo rule turns on likeness, not on how the image was made. The company has said publicly that it recommends authenticity, permits tools including AI to enhance or create a profile photo, and requires the result to reflect your likeness — and that it can remove photos that don't. That's the same standard it has long used to pull stock images, cartoon characters and other people's faces. So there's no AI-photo ban to fall foul of, and no public detection tool sweeping profile pictures. There's a rule, and it's about identity. The practical test is whether a recruiter who had met you would recognise the photo.

In practice the line sits here. Fine: an AI headshot trained on your own selfies that still looks like you, AI retouching applied to a real photo, an illustrated portrait of yourself. Not fine: a fabricated face, a jaw or hairline or decade reshaped past recognition, somebody else's image. LinkedIn's broader synthetic-media policy also covers manipulated depictions of real people, which gives it grounds to act on an impersonating picture. Enforcement is reactive — profiles get examined when they're reported, not swept in advance — so nobody stopped me tells you nothing about whether you're compliant. If you want the full argument, here's where we draw the line on synthetic portraits.

Don't plan around provenance labels, either. Platforms re-encode almost every image at upload, and the C2PA content-credentials manifest that carries provenance data lives in the part of a file that re-encoding tends to destroy — the standard's own threat model concedes that a social upload, a screenshot or a format conversion can remove it. Reports conflict on exactly what LinkedIn preserves and displays. A pixel-level watermark like SynthID survives that pipeline far better than metadata does, which is precisely why Google built one. The takeaway for you is blunt: don't assume a label will out you, and don't assume stripping metadata protects you. Build a profile that holds up when a human looks at it twice.

Will an ATS flag your CV as AI-written?

No. Mainstream applicant tracking systems parse and rank; they don't adjudicate authorship. Reviews across the big platforms — Workday, Greenhouse, Lever, Taleo, iCIMS, SmartRecruiters, Jobvite and the rest — have repeatedly found no built-in AI-text detection. What their AI actually does is extract your text and score it against the posting on titles, skills and keywords. The one documented exception is Oracle Recruiting, whose 2026 release notes describe an opt-in screening agent that scores applicants for authenticity, including AI-generated candidates. Plenty of employers also run CVs through standalone detectors, but those sit outside the ATS pipeline, and they're notoriously shaky on documents as short and formulaic as a CV.

Vendors avoid authorship detection for a boring, rational reason: the false positives would be brutal. Detectors routinely flag non-native English writers and anyone who ran a grammar checker over a draft, and at enterprise volume even a one-in-a-hundred error rate means thousands of real candidates wrongly tagged. No ATS vendor wants that argument, let alone that lawsuit. Where flagging layers do exist, they tend to push a candidate into a lower-priority queue rather than reject outright. Text watermarking is a separate thread worth understanding — we've unpacked what the ChatGPT text watermark does inside a recruiter's ATS — but the short version is that it isn't what gets CVs binned.

The screen that actually catches AI writing is a person. Roughly 88% of hiring managers say they can tell when a candidate used AI on a CV or cover letter, and what they're responding to is texture, not telemetry: verbs with no numbers attached, three bullets saying the same thing in different words, spearheaded cross-functional initiatives to drive synergies, a tone that sounds nothing like your interview answers. Generic phrasing costs you mechanically too, because it fails to mirror the specific terminology in the job ad — which is exactly what the parser weights. Vague AI prose loses twice over: once to the ranking algorithm, once to the human who reads what survives.

How do you make your CV read human before you send it?

Write for specificity, not for a detector. The fix is almost always numbers: team size, budget, percentage, timeframe, tool name. Improved onboarding is AI-shaped; cut new-hire ramp time from six weeks to four across 40 hires is yours and nobody else's. Strip out the adjectives — dynamic, results-driven, passionate — and keep the nouns a practitioner would use: the actual SIEM you ran, the exact framework version, the ticketing system, the quota and your attainment against it. Read every bullet aloud; if you'd never say it out loud to a colleague, rewrite it. Use AI for structure and first drafts, then put your own vocabulary back in, line by line.

Then check the mechanics, because formatting sinks more applications than wording does. Single column. Real section headings an extractor recognises — Experience, not Where I've Made an Impact. One date format throughout. No icons or text boxes carrying load-bearing information, and no skills buried in a sidebar that the parser reads out of order. Make the CV and the LinkedIn profile agree on titles and dates, because mismatches read as carelessness at best and invention at worst. And if you're specifically worried the draft sounds machine-made, run it through a check for AI-sounding phrasing instead of guessing from a gut feeling.

This is where automated feedback earns its keep. A free CV analysis on HRLens takes a PDF, a Word file or even a photo of your CV — Hebrew OCR included — and returns a score out of 100 alongside five category scores covering experience, tech stack, impact and ownership, clarity and structure, and ATS compatibility, plus your strengths and a visual read on the layout. Paste in the job description and it lists the skills you're missing for that specific role. The $3.99 full analysis adds the fix list, the why you might be rejected breakdown and a hiring-probability estimate. That's a concrete read on the document recruiters will open — far more useful than uploading your face to a watermark checker and hoping.

Frequently asked questions