Home / Guides

Check a PDF or Word doc for AI?

Upload the file and our checker reads it on your device, then scores it like pasted text. Here is what happens to your document, what scanned PDFs need, and why the score means the same as always.

Updated October 31, 2026 · By the IsThisAI? team

The short answer

Yes. Upload the file to our checker and it reads the text on your device, then scores it like pasted text. A PDF or Word document is just a container. What matters is the words inside.

How it works here

Our checker accepts .txt, .docx, and .pdf files. When you upload one, your browser reads the text out of the file locally. Nothing is uploaded to a server. The extracted text is then run through the same two open-source detection models as pasted text, plus the same style checks.

A few practical notes:

• Scanned PDFs need readable text. If the PDF is photos of pages rather than real text, there is nothing to extract. You would need OCR first, and OCR errors can themselves look strange to a detector.
• Formatting is stripped. Headings, tables, and footnotes become plain text. That is fine for scoring, but very table-heavy documents give the detector less prose to work with.
• Long documents are split. Text is scored in chunks and the scores are averaged by length, so a 50-page report gets the same treatment as a one-page letter.

What to watch for in documents

Documents carry quirks that pasted text does not. Tracked changes, comments, and metadata can reveal editing history that is more informative than any detector score. In Word, the document properties sometimes show author names and edit time. None of that proves AI use, but it is real evidence, which is more than a score gives you.

Also note that many real documents are hybrids: a person writes a draft, AI polishes it, a person edits again. Detectors see the final blend and report a single number. A mid-range score on a document often means exactly that kind of mixed process, not a simple yes or no.

Accuracy on documents vs. pasted text

There is no meaningful difference. The detector scores the extracted words the same way either way. What changes the accuracy is the kind of writing: long, specific, idiosyncratic prose scores most reliably, while short, formal, or template-driven documents score least reliably. A fill-in-the-blanks report template will look AI-like no matter who filled it in.

Bottom line

Checking a file is no different from checking pasted text, and the same limits apply. Read the score as a probability, look at the actual writing, and never treat the number as proof of anything about anyone.

Step by step: checking a document

The process is the same as checking pasted text:

• 1. Open the checker and click "Upload file" (or drag the file onto the page).
• 2. Choose your .txt, .docx, or .pdf file. It is read on your device; nothing uploads.
• 3. The text is extracted and scored. Long documents are split into chunks, each chunk is scored, and the results are averaged by length.
• 4. Read the score the same way you would for pasted text: under 30% likely human, 30 to 80% uncertain, over 80% strongly AI-like. And remember it is a probability, not proof.

Scanned PDFs and photos of pages

If the PDF contains images of pages instead of real text, there is nothing for the checker to extract. You will need to run OCR (optical character recognition) first, which turns the images into text. Two cautions: OCR introduces errors, especially on unusual fonts or low-quality scans, and those errors add noise that can push the score around. Also, a scanned document tells you something on its own: most AI writing workflows produce clean digital text, so a scanned handwritten or heavily marked-up page is already evidence of a human process.

Document metadata: evidence detectors cannot give you

Files carry history that scores do not. A Word document's properties can show the author name, the software used, total editing time, and revision counts. A PDF's metadata shows creation and modification dates and the producing application. None of this proves who wrote the words, but it is concrete evidence about the document's life, which is more than a probability score provides. In a dispute, "this file was created yesterday in one 20-minute session" is a fact you can discuss. "The detector said 82%" is not.

Sources

1. Weber-Wulff et al., "Testing of detection tools for AI-generated text," International Journal for Educational Integrity 19, 26 (2023). https://link.springer.com/article/10.1007/s40979-023-00146-z
2. IsThisAI? About page: how our detector works. https://getisthisai.com/about/

This guide is general information, not legal or hiring advice. AI detection results, including ours, can be wrong.