Free Download · No Signup

Free Sample Documents for Testing

50 realistic synthetic documents — invoices, letters, forms, emails, receipts, agreements and more — ready to feed into any OCR, extraction, or document-AI pipeline. Every entity is fabricated, so there's nothing to redact and nothing to leak.

What's inside

• 10+ document types — invoices, business letters, memos, emails, receipts, agreements, forms, reports, tables, and more.
• Clean / degraded pairs — the same documents at full quality and with realistic OCR degradation, so you can measure exactly what noise costs your pipeline.
• Corrupt-text-layer examples — PDFs that look clean but carry a corrupted embedded text layer, exposing tools that read embedded text instead of running real OCR.
• Native formats — real .docx, .xlsx, .eml, .csv, and .txt files alongside PDFs.
• Zero real PII — structurally invalid SSNs, reserved 555-01xx phone numbers, .example email domains, fabricated companies and addresses.

What people use them for

• Smoke-testing OCR and extraction pipelines before writing a data-handling review.
• Benchmarking cloud OCR services (Textract, Azure Document Intelligence, Google Document AI) — see our benchmarking guide.
• Testing DLP scanners, redaction tools, and eDiscovery processing.
• Populating demo environments and sales-engineering sandboxes.

Need more than 50?

The full desktop app generates unlimited documents across 33 types with a 0–100% degradation slider, seeded reproducible output, ground-truth manifests, and ML-ready COCO / LayoutLM / FUNSD / DocVQA annotation exports — $199 one-time, runs fully offline on Windows and Linux. See pricing.

Frequently asked questions

Are these documents free to use commercially?

Yes. The samples are synthetic and free to use for testing, development, demos, and evaluation. No signup or attribution required.

Is any of the data real?

No. Every name, address, company, SSN, and dollar figure is fabricated — enforced by the generator's test suite.

Can I upload these to cloud OCR services?

Yes. Because no real entities appear anywhere, they're safe to send to any third-party API with zero PII exposure.