Checking PDF, DOCX and HWP documents written with AI

Reports, cover letters and assignments usually travel as document files. A document holds traces of which program made it and how, alongside the body text, so there is a little more to look at than with pasted text.

The text detector reads the body

Body text is extracted from PDF, DOCX, HWP and HWPX and read by the text detector. In quote-heavy documents the author's own words are read again with quotes removed, so other people's text does not sway the result. Text detection is in beta, and that holds for documents too.

Document structure traces

DOCX revision identifiers, the authoring application record and the PDF producer say how a document came to be. A document pasted in one go and one revised over time differ in these traces, but they alone never decide AI authorship.

Problems specific to PDF

PDF line breaks and columns can split sentences on extraction. Origma rejoins broken lines before reading, but a scanned PDF whose letters are images has no readable text and comes back inconclusive.

How to use the result

Document results show text detection and structure traces separately. In grading or hiring, do not conclude from the result alone; use it to start a conversation that also covers the writing process and drafts.

Frequently asked

Are HWP files supported?
Yes. Both HWP and HWPX are read for text and structure.
Can a human document be called AI?
Yes. Some human documents were flagged in our document gate, and we publish how many.
What about documents full of tables and images?
With little text to read there is less to judge, so inconclusive results increase.
Do you check whether the file itself was made by AI?
The authoring program and revision history are read, but they are clues about the writing process, not direct evidence of AI use.

Read next

This is general guidance, not legal advice. A scan result is a probabilistic assessment, not proof.

Try a file