Human or AI checker
In an era where synthetic media and generative writing are ubiquitous, distinguishing genuine human work from machine generation is essential. This tool provides an unbiased, evidence-backed verdict on any submission.
Rather than giving a black-box percentage, we inspect embedded camera EXIF data and C2PA Content Credentials first, followed by multi-model neural vision and language analysis.
The order matters. Provenance data is something the file states about itself, so when it is present the answer is close to settled. Statistical detection is a guess about pixels or sentence rhythm, and it is only consulted when nothing better survives. Every result names which layer produced it, so you can see whether you are looking at evidence or at inference.
Be realistic about what a verdict is worth. Detection models are trained on particular generators and lose accuracy on anything newer, and a screenshot strips the metadata that carries the real signal. A confident answer here is worth acting on; an uncertain one genuinely means the file does not contain enough to decide, which is the honest outcome more often than most tools admit.
Drag a file here, or
JPEG, PNG, WebP, AVIF, PDF or .txt · up to 10 MB · you can also paste a screenshot
By uploading you agree to our Terms and Privacy Policy. Your file, IP address, approximate location and device details are stored.
What this checks
- Cryptographic provenance check (C2PA, IPTC, and camera hardware EXIF)
- Neural network image artifact classification for photographic vs synthetic media
- Linguistic burstiness and perplexity analysis for human vs synthetic text
- Absence of authentic human camera sensor noise and lens distortion
- Zero-knowledge privacy — no account needed, private processing
Frequently asked questions
What gives the most conclusive proof that content is human?
For images, verifiable camera sensor metadata, lens profiles, exposure settings, or cryptographic C2PA provenance prove real-world capture. For text, natural bursts of varying sentence length and idiomatic expressions indicate human authorship.
How does the tool handle hybrid human-AI content?
When content has been co-created (e.g. human outline expanded by AI, or AI photo edited in Photoshop), the evidence will show mixed signals, and our report reflects this nuanced reality.
Why does a file sometimes come back as 'not enough evidence'?
Because that is frequently the truthful answer. Social platforms strip metadata on upload, screenshots discard it entirely, and re-encoding destroys the statistical traces detectors rely on. When a file arrives stripped of both, there is nothing left that separates human from synthetic, and inventing a confident number would be dishonest rather than helpful.
Can I use this result as proof in a dispute?
No. Treat it as a reason to ask questions, never as evidence on its own. No detector on the market is accurate enough to support an accusation, and text detection in particular produces more false positives for people writing in a second language — so the cost of a wrong answer falls unevenly. Corroborate independently before acting on anything.
Does checking a file here make it public?
No. Uploads are stored privately and result pages are excluded from search engines. Anyone you send a result link to can open it, so share links only with people you intend to see the file, and you can request removal at any time.