Focused on one job: deciding whether a document was produced by a generative AI model.
Upload a file to start verification.
Drag & drop your document
or click to browse
PNG, JPG, WebP, PDF · Max 10MB
Your file stays in your browser until you create an account and confirm your email. Originals are deleted after analysis.
Estimated savings based on replacing a 10–15 minute manual document review with automated TrueDoc analysis.
Proprietary detection scans template variance, metadata drift, pixel-level retouching, and structural anomalies the human eye misses.
Inspects EXIF, software signatures, edit history, and structural fingerprints.
Originals are processed in encrypted memory and removed after analysis. Reports stay redacted by default.
Generative models can now produce convincing IDs, invoices, statements, contracts, and screenshots in seconds. Template-based 'fake document' detectors miss them entirely.
TrueDoc's AI-generation detection runs as a first-class signal layer, surfacing fingerprints that current generative models leave behind.
Fully synthetic IDs and passports from generative image models
AI-generated bank statements and pay stubs
AI-drafted invoices and contracts that reference no real transaction
AI-generated screenshots of UIs that never existed
Documents passed through generative 'restoration' that hides edits
PDF or image, up to 10MB.
Latent fingerprints, statistical signatures, and structural cues.
Cross-checked against ELA, metadata, math, and layout signals.
AI-generation probability plus the full forensic evidence map.
Most "AI detector" tools look at one signal — usually a perplexity score on extracted text. The ai generated document detector runs that as one layer of many. It also evaluates metadata lineage (software, edit history, geo), pixel-level forensics (ELA, font kerning, retouching regions), and structural anomalies in the underlying PDF or image container.
The reason: fully synthetic ids and passports from generative image models rarely leaves only one fingerprint. A convincing forgery usually fails on two or three of those layers, even when one of them looks clean.
A high-risk verdict on Identity documents, Financial documents (statements, pay stubs, invoices, receipts), Contracts and supporting documents returns per-field evidence — not just a score. You see the suspicious regions highlighted on the page, the specific metadata fields that triggered the flag (for example, "AI-generated bank statements and pay stubs"), and the layer each finding came from.
That structure is what makes the verdict actionable: fraud and risk teams can read why a document was flagged before deciding to reject, request a reupload, or escalate.
Scanned originals, mobile-camera shots, and re-exported PDFs are the three most common sources of benign anomalies. The ai generated document detector scores those differently from the patterns associated with deliberate forgery — for example, a recompressed JPEG from a phone is not treated the same as a recompressed JPEG with a font substitution.
When a document is flagged, the report tells you which signal triggered it. If the only signal is a low-confidence compression artifact, the verdict is downgraded rather than counted as fraud.
No credit card. Redacted report in under a minute.