White paper · October 2026

The Forger's Toolkit

How receipts and claim documents are faked today, what each technique leaves behind, and how to check for it without burying reviewers in false positives.

Free download

Five ways documents are faked, and what gives each one away.

Editors, phone apps and AI image generators have made a convincing fake invoice or receipt a few minutes' work. This paper sorts what fraudsters actually do into five techniques, sets out the trace each one leaves in the file, the pixels or the numbers, and reports how well checks for those traces work on public test data, including where they don't.

Six pages, with every industry figure sourced
Measured on 6,760 documents from public test sets
What to check first, for claims and expense teams

Download the white paper (PDF, 0.9 MB)Book a demo

The white paper's cover: 6,760 receipts, invoices, statements and photos plotted by the risk score Assaya gave them, genuine ones clustered near zero and forged ones spread across the range. — The cover plots every document from Assaya's public-dataset tests at its risk score.

Inside

The five techniques at a glance.

1. Edit the real document

Opens a genuine PDF invoice or estimate in an editor and types a new total over the old one.

What it leaves behindContent added to the page, the old value still in the file, an editor in the file's history.

2. Edit the picture

Changes digits on a photographed receipt, or paints damage onto a photo.

What it leaves behindThe editing program's signature, cloned pixels, characters that don't match their neighbours.

3. Make it from scratch

Uses a receipt-generator website, a template, or an AI image generator.

What it leaves behindCharacters too perfect to have been printed, signed AI provenance, figures that don't add up.

4. Reuse a real one

Claims the same receipt twice, a colleague's, one from before the trip, or one larger than the card charge.

What it leaves behindThe same purchase on another claim, a date outside the trip, a total that doesn't match the card.

5. Fool the reviewer's AI

Hides instructions in the document for the language model that extracts or reviews it.

What it leaves behindInvisible or white text, text addressed to an AI.

Also in the paper: separating fraud from ordinary paperwork, what we recommend, and how the measurements were made. Download the PDF

Proof before payout.

See how Assaya highlights suspicious findings, explains what triggered them and gives reviewers what they need to take the next step. A focused 30-minute walkthrough; no sensitive documents required.

See Assaya in action