Document checking

How to check a PDF for AI

A document contains more than running text: headings, tables, text boxes, images, metadata and sometimes scanned pages. A reliable check therefore starts with trustworthy extraction, not with the final score. A detector signal is a reason to review, never proof of authorship.

1. Use the best source file

Prefer the original PDF or DOCX. A scan may have no usable text layer; an exported PDF can scatter words across separate text boxes. After upload, confirm the filename, word count, section headings and preview text look right before reading any result.

2. Read sections separately

An average over the whole document can hide a conspicuous passage. Review the signal and text per section. A short title, address list or table is too little natural language for a meaningful style judgement and should not outweigh several full paragraphs.

3. Combine signals

• Compare tone and terminology between sections.
• Check embedded images separately when they are content-relevant.
• Ask for drafts, references or revision history.
• Keep the downloadable detection report as a record of the check, not as an authenticity certificate.

4. Use the document detector

Open the AI document detector. It accepts PDF, DOCX, HTML, TXT and Markdown, extracts readable sections locally in the browser, and checks supported embedded images separately. For single passages the text detector fits better; for imagery use the image detector.

What the result does and does not say

A strong signal means: this document carries patterns common in synthetic text. That is a reason to collect source material and have a second person review — not to accuse or reject anyone on the score alone. A low signal proves nothing either; new generators, editing, translation and templates all change detector behaviour. Read the measured limits and the other guides before drawing conclusions.