PDF Forensics: Detect Edited & Fake PDFs

Veryfi PDF Forensics reads every layer inside a bank statement, pay stub, or invoice: text, fonts, metadata, digital signatures, and file structure. It checks them against 110 fraud signals in under 3 seconds and returns a risk score, the evidence, and a plain-English explanation before the document enters your workflow.

Get Started for Free Free Demo
PDF Forensics: Detect Edited & Fake PDFs

Part of the Fraud Detection Suite

These checks run with the rest of Veryfi Fraud Detection and come back in the same response, under meta.fraud: one color, a score, and the reasons.

See every fraud check
  • AI-Generated Documents

    Spots receipts, invoices, and checks made with ChatGPT, Midjourney, Stable Diffusion, DALL-E, or GANs.

    You are here
  • PDF Forensics

    Reads inside the file: the app that made it, its fonts, and text laid over the original. Flags edited and AI-generated PDFs.

    You are here
  • Image Forensics

    Reads the layers inside a photo: error levels, camera noise, copied pixels, and EXIF data. Flags edited, AI-generated, and recaptured photos.

    You are here

How Veryfi reads the layers inside a PDF and uses multi-modal AI to detect fraud

A PDF bank statement looks like one page, but Veryfi reads it as eleven layers: the rendered page, annotations, text objects, extracted text, fonts, images, a page thumbnail, metadata, a digital signature, and the file structure, and then a multi-modal AI model reads the pixels, text, and structure together. In this example the ending balance was edited. The page looks normal, but nine layers show the edit: a white annotation box over the old number, a new text object drawn on top, two balances in the extracted text, a font the bank never used, a thumbnail that still shows the original value, metadata from an online PDF editor, bytes added after the digital signature, an incremental update that leaves the original content stream in the file, and an AI model that scores the statement 2% likely to be AI-generated but 96% likely to be edited at the ending balance. Veryfi Fraud Detection reads every layer, not just the pixels, and uses AI to tell a generated document from an edited one.

AI PDF forensics

The page looks real. The layers say otherwise.

Veryfi Inspect X-Ray checks every PDF against 110 fraud signals

A PDF goes into Veryfi Inspect X-Ray and is read against 110 fraud signals in three families. File and origin (34 signals): 14 file signals, 8 metadata signals, and 12 timeline and duplicate signals. Pixels and AI (25 signals): 16 pixel signals and 9 AI-generated and screen recapture signals. Text and PDF (51 signals): 9 OCR signals, 26 PDF structure signals, and 16 PDF content signals. On top of these, 14 document integrity checks confirm the file is complete and consistent. Out come three results: the extracted data (vendor, invoice number, date, and total), the fraud signals that fired (a text layer edited after signing, a font mismatch on the total, and metadata from an online editor), and an insight (a risk score of 92 out of 100, so the document is held for review).

Vee, the Veryfi Inspect advisor, explains every fraud signal on a document and what to do next

Vee Advisor

Every signal, explained by Vee.

  1. Hi, I’m Vee. I read this file against all 110 Inspect X-Ray signals. Let me walk you through what fired, why it matters, and what to do next.
  2. Scan Signals: 83% risk. 9 of 110 signals fired across six families, and they hit the fields that matter most. That’s high risk.
  3. Verdict: reject as proof of address. I’m confident on the call: the document type alone rules it out. Fax OCR is only 76%, so I flag details instead of guessing.
  4. 1. Wrong document type. This is a Plumbus Supply Co. credit memo plus a letter from Rick. Proof of address has to come from the issuer, like a utility bill or bank statement.
  5. The page types give it away. Page 1 reads as a credit note (92%) and page 2 as other (78%). Neither is the kind of document this workflow expects.
  6. 2. Address block size anomalies. Morty’s street line and city line are set at different sizes (CV 43%). The address is the field fraudsters edit most.
  7. Zoom in on the address. “137 Citadel Way, apt C-137” is smaller and sits off the city line’s baseline. Uneven size or spacing points to an edit. Uniform blur would just be fax noise.
  8. 3. Too old. The memo and Rick’s letter are both dated April 1, 2025. Yes, April Fools’ Day. Proof of address must be dated within the last 3 months.
  9. Here’s the date. April 1, 2025 is 19 months before this check ran. Even a perfect document this old would fail a KYC address check.
  10. Dates span 3.2 years. The fax header’s Thursday matches its date, so the dates are consistent, just stale. The newest one is 16 months outside the window.
  11. Arithmetic: not checkable. The line items add up to 4,206.90 schmeckles, but there’s no printed total to compare against. I can’t confirm the math or flag it.
  12. Got questions? Chat with me. Ask anything about this document. I answer from its X-ray report, like whether that address is really just fax noise.
  13. That’s every signal. 3 findings: 2 high, 1 medium. Reject it as proof of address and ask Morty for the issuer’s own PDF. Got more questions? Just ask me.
Fused JSON Example
“meta”: {
“duplicates”: [
{
“id”: 191347946,
“score”: 1,
“url”: “https://scdn.veryfi.com/receipts/…”
}
],
“fraud”: {
“attribution”: “High velocity, Handwritten characters, LCD photo, Digital background, Duplicate receipt”,
“color”: “red”,
“decision”: “Fraud”,
“digital_tampering_fields”: [
“total”,
“line_items.10.total”,
“line_items.14.total”
],
“fraudulent_pdf”: {
“font_mismatch”: 0,
“fraudulent_pdf_creator”: 0,
“score”: 0.94,
“text_overlay”: 0.94
}
}
...
Cont… Fused JSON Example
...
“images”: [
{
“is_lcd”: true,
“score”: 0.96
}
],
“score”: 1,
“types”: [“handwritten characters”, “LCD photo”, “screenshot”, “duplicate”, “fraudulent pdf”, “generated document”]
},
“handwritten_fields”: [
“line_items.0.total”,
“line_items.1.total”
],
“warnings”: [
{
“message”: “Line item #14 has a problem. Quantity x Price != Total”,
“type”: “line_item_amount_missmatch”
}
]
}

Developer Friendly API (Headless Option)

PDF fraud signals are baked into the same JSON response as your extracted data API. 1 API. No second API call. No extra integration. AI that reads documents and knows when they’re lying.

Most teams are up and running in an afternoon.

color is your routing logic.
attribution is your audit trail.
digital_tampering_fields pins exactly which values were touched.

Thresholds ship with sensible defaults. Tighten or loosen per signal whenever you’re ready.

Use Cases

Anywhere a PDF decides who gets a loan, an account, or a payment. See how Veryfi reads the layers inside a PDF.

Frequently Asked Questions for PDF Forensics

  • What is PDF forensics?

    It means checking the inside of a PDF (text objects, fonts, metadata, signatures, revision history) to find out whether the file was changed after the issuer created it. Looking only at the page can’t show this.

  • How is this different from OCR or a visual check?

    OCR reads what the page says. Forensics checks whether what it says is original. An edited PDF can look perfect and still carry a second text layer, a new font, or an editor’s name in its metadata.

  • Which documents does it work on?

    All types of documents that are of format type PDF.

  • Does it work on scanned PDFs, or only original digital ones?

    All types. Those generated by industrial scanners during back-office digitization, created by humans using PDF creator like tools, those FAXed and more. As long as it’s a PDF we got you.

  • Can it tell an AI-generated PDF from an edited one?

    Yes. The multi-modal model scores both separately, for example “2% AI-generated, 96% edited at the ending balance.”

  • Which tools it catches?

    Online PDF editors, Acrobat “Edit text”, files exported from Word or Excel, template-farm documents, and PDFs generated by ChatGPT or other AI.

  • What happens if a PDF is digitally signed?

    Veryfi checks which bytes the signature covers and flags anything added after signing.

  • Will it flag real documents just because they were re-saved or compressed?

    Not on its own. One signal changes the score, but red needs several signals pointing to the same field. Yellow sends the file to a person for review.

  • Can fraudsters get around it by removing metadata?

    Removing metadata is itself a signal, and the other layers (fonts, text objects, file structure, pixels) still carry traces.

  • Do I have to set up rules or train a model?

    No. It works on the first day, and you choose what green, yellow, and red do in your workflow.

Fraud is live 24×7

Intercept Document Fraud

Intercept forged IDs, padded expenses, and fake invoices flooding KYC/KYB, Travel & Expense, Bill Pay, Audit and other areas where fraud is a risk. Stand up Veryfi in minutes and shut them out before they clear.

No credit card required. Live in minutes.