Document Forensics

AI-Generated Document Detector

Focused on one job: deciding whether a document was produced by a generative AI model.

Fraud and risk teamsKYC and onboarding teamsHR reviewing applicant documentsAnyone vetting a suspicious file
4.9·132+ reviews

Check a document instantly

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.

Advanced Verification
Originals Deleted
Redacted Reports

TrueDoc ROI and performance stats

<120sForensic report
10× fasterFraud review
40+Fraud signals
8+ hrsSaved / 100 docs

Estimated savings based on replacing a 10–15 minute manual document review with automated TrueDoc analysis.

Built on Trusted AI Infrastructure
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic
Google Cloud
Gemini
OpenAI
Anthropic

Multi-layer forensic logic

Proprietary detection scans template variance, metadata drift, pixel-level retouching, and structural anomalies the human eye misses.

▸ Document Analysis · LiveID: 8829-XQ
Risk score: High · 94%Signals matched: 12,042

Metadata deep-dive

Inspects EXIF, software signatures, edit history, and structural fingerprints.

SoftwareAdobe Photoshop 2024
ModifiedDetected
Geo-tagMismatch

Privacy-first by design

Originals are processed in encrypted memory and removed after analysis. Reports stay redacted by default.

No training on your data
Team & admin controls
▸ 01 · The Problem

Why eyeballing a document no longer works

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.

▸ 02 · Fraud Signals

What we look for

Cross-checked across 5+ vectors
▸ Primary signal

Fully synthetic IDs and passports from generative image models

Detected at pixel + metadata + structural layers

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

What gets checked

Identity documents
Financial documents (statements, pay stubs, invoices, receipts)
Contracts and supporting documents
Screenshots and image proofs
▸ 03 · Workflow

From upload to verdict

01

Upload the document

PDF or image, up to 10MB.

02

Run AI-generation analysis

Latent fingerprints, statistical signatures, and structural cues.

03

Combine with forensics

Cross-checked against ELA, metadata, math, and layout signals.

04

Read the verdict

AI-generation probability plus the full forensic evidence map.

How the ai generated document detector differs from a generic AI check

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.

What a high-risk report actually shows

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.

Common false positives and how we suppress them

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.

Run a real document. Get a forensic verdict.

No credit card. Redacted report in under a minute.

▸ FAQ

Frequently asked questions