# How to check a PDF for AI-generated text

> Check a paper or thesis PDF for AI writing without copy-paste errors: what to upload, what to leave out, and how to read the result page by page.

Updated 2026-11-03. 5 min read.

Checking a PDF for AI writing goes wrong most often before the detector even runs: text copied out of a PDF is broken, mixed with headers and references, and cut off by character limits. Here is how to check a paper or thesis PDF so the result reflects the writing, and how to read what comes back.

## Why copy and paste gives the wrong result

- Line breaks split words ("detec- tion") and two-column layouts interleave sentences from both columns.
- Running headers, page numbers and footnotes land in the middle of paragraphs.
- The reference list and figure captions come along, and formulaic text like that reads as machine-written to any detector.
- Many free checkers stop after a few thousand characters, so only the first pages are judged.

## Prepare the PDF

1. **Use a text PDF.** Export from Word, LaTeX or Google Docs. A scanned PDF is a picture of text and cannot be checked without OCR, which adds its own errors.
2. **Check the version.** Use the exact version you will submit, so the result applies to what readers will see.
3. **Mind the size.** If figures make the file large, export a copy with compressed images or without figures; only the text matters.
4. **Split a long thesis.** Check chapters as separate PDFs if your detector limits the words per check.

## Run the check

1. **Upload the PDF.** Use a detector that reads the file directly rather than one that wants pasted text.
2. **Leave out what says nothing about authorship.** References, captions, tables and headings should not be scored. Good academic detectors drop them for you.
3. **Note the coverage.** Check how much of the paper was actually scored, and whether anything was skipped.

## Read the result page by page

A single percentage tells you little. Open each flagged passage at its page and ask whether it is generic: a sentence that could sit in any paper on the topic. Those are worth rewriting with your data, your reasoning and your field’s terms, whatever produced them.

Expect some flags on methods text, standard definitions and boilerplate statements; they read alike in every paper. A detector score is a reading guide, not proof of who wrote the text.

## Protect an unpublished paper

- Read the tool’s data policy before uploading a manuscript under review, a thesis before deposit, or anything with confidential data.
- Prefer tools that do not keep the PDF and that show results only to you.
- Check whether your institution or the data-sharing agreement for your study allows third-party processing.

## Do this faster with NoveltyAxis

Reads the PDF itself, leaves out references, captions and tables, and shows each flagged passage with its page. The PDF is never stored. The free plan includes one check a month. See the [ai detector](https://noveltyaxis.com/tools/ai-detector/index.md).

## Questions

### Can you check a PDF for AI writing for free?

Yes, with limits. Free tiers usually cap the characters or checks; the NoveltyAxis AI Detector includes one full-paper check a month on the free plan, showing the share of the paper that reads as AI-written.

### Can AI detectors read scanned PDFs?

Usually not. A scanned PDF holds images of text. Export the original document as a PDF instead, or run OCR first and expect some errors in the result.

### Should the reference list be checked for AI?

No. References are formulaic and say nothing about who wrote the paper; including them makes a score less meaningful. Check instead that every reference is real.

### Is a high AI score proof of misconduct?

No. Detectors flag some human writing, especially formal or formulaic text. Treat the score as a pointer to passages worth rereading.

HTML version: https://noveltyaxis.com/guides/how-to-check-a-pdf-for-ai/
