PDF AI Detector
Drop in a PDF and check its writing for AI. The file is opened on your own device and never uploaded; you get a 0–100 score, the signals behind it and the phrases pulling it up.
To check a PDF for AI, import the file and score the text inside it. Humanit opens the PDF in your browser without uploading it, rates the prose 0–100 and marks the phrases that read as machine-written. Treat the score as a signal, not a verdict.
Quick facts
- Files
- PDF, text-based or scanned. Word, ODT, RTF, TXT and photos work too.
- Is the file uploaded?
- No. The PDF is opened inside your browser; only its text goes into the editor.
- Words per check
- 300 words a check without an account, 1,500 with a free one, up to 6,000 on paid plans.
- Scanned pages
- Read with OCR on a free account: 10 pages a day, 100 on paid plans.
- File limits
- 20 MB per file, up to 30 pages or photos at once.
- Cost
- Free. The detector never spends words, with or without an account.
- What you get
- A 0–100 score, ten sub-signals, the phrases that read most like AI and suggested fixes.
How it works
- 1Drop the PDF on the box. Or choose Import a PDF. Text pages are read instantly in your browser; scanned pages are read with OCR once you have a free account.
- 2Trim it to the prose. Delete the title page, repeated headers and the reference list, so the score reflects the writing and not the paperwork.
- 3Run Detect AI. Up to 300 words with no account, 1,500 with a free one. Longer PDFs are checked a section at a time.
- 4Read the breakdown. Look past the headline number to the sub-signals and the highlighted phrases. That is where the useful information is.
What happens to your PDF when you drop it in
Nothing leaves your computer at that point. The file is opened by Mozilla's PDF.js, the same open-source reader Firefox uses to display PDFs, running inside this page. It pulls out the text layer, re-joins lines that were wrapped at the page edge, heals words that were hyphenated across a line break and drops bare page numbers. Only that cleaned text is placed in the editor, where you can read and trim it. The text is sent for scoring when you press Detect AI, and not before. If you are signed in, the check appears in your history, where you can delete it.
Text PDFs and scanned PDFs are different files
A PDF exported from Word, Google Docs or a journal website carries a text layer: the words are stored as characters, which is why you can select and copy them. A PDF made by a scanner or a phone camera usually stores a picture of each page instead, with no characters at all. The importer checks every page. Pages with a text layer are read on the spot; pages that are only an image are sent to text recognition, which needs a free account and counts against a daily allowance of 10 pages. A mixed document, such as a typed essay with a scanned appendix, is handled page by page. The quickest way to tell which kind you have is to try selecting a sentence in your PDF viewer. If the cursor will not highlight words, it is a scan.
Saving as a PDF does not hide AI writing
It is a common belief and it is wrong. Exporting to PDF changes the container, not the words. The text layer holds exactly the sentences you wrote, and every checker that accepts PDFs, Turnitin included, reads that layer. The only way to make a PDF opaque to a checker is to flatten it into images, and a submission nobody can read draws more questions, not fewer. If a PDF reads as AI-written, the fix is in the writing. Use the PDF humanizer to rewrite flagged passages, or rework them by hand, then check again.
Score the prose, not the paperwork
A PDF contains a lot of text that is not prose: a title page, running headers that repeat on every page, figure captions, a reference list, appendix tables. Formulaic text like that scores oddly in either direction and drags the result away from what you actually want to know. Before you run the check, delete everything that is not continuous writing. Then work in sections. With a free account the detector takes 1,500 words at a time, which is about three single-spaced or six double-spaced pages, and checking the introduction, the body and the conclusion separately tells you far more than one number for the whole file. The section that scores highest is where to look first.
How to read the result
The headline figure is a probability-style score from 0 to 100, where higher means the text shares more of the statistical habits of machine-written prose. Under it sit ten sub-signals, among them perplexity (how predictable the word choices are), burstiness (how much sentence length varies), AI vocabulary, discourse markers and personal voice. Below those are the phrases pulling the score up most and a few suggested fixes. The phrases are the most useful part, because they point to the sentences to look at. No detector is a lie detector. Formal academic writing and writing by non-native English speakers are flagged more often than they should be; a 2023 study published in Patterns found that more than 61% of genuine TOEFL essays were misclassified as AI-written by the detectors it tested.
Checking a PDF someone else wrote
Teachers, editors and hiring managers use this page too. If a submission scores high, treat that as a reason to look closer, not as a finding. Compare the writing with earlier work by the same person, ask for drafts or version history, and talk to them. A score on its own cannot tell you who wrote something, and the cost of a false accusation is far higher than the cost of a second conversation. Your institution may also have rules about uploading student work to outside tools; here the file is read locally, but the text you choose to score is sent for analysis, so check your policy first.
Who it's for
| In your PDF | What the import does | Before you check it |
|---|---|---|
| Text on ordinary pages | Read in your browser. Lines that wrap at the page edge are re-joined into whole paragraphs. | Nothing to do. This is the text that gets scored. |
| Page-number lines ("12", "Page 3 of 10") | Removed automatically. | Nothing to do. |
| Running headers and footers | Kept as text, once for every page they appear on. | Delete them. Repeated title lines are not prose and skew the score. |
| Footnotes | Imported where they sit on the page, usually between paragraphs. | Delete them unless you want them scored with the paragraph. |
| Reference list or bibliography | Imported like any other text. | Delete it. Citation lists score unpredictably and are not your prose. |
| Scanned pages with no selectable text | Read with OCR. Needs a free account; 10 scan pages a day free, 100 on paid plans. | Sign in first; a free account reads scanned pages. |
| Two-column layouts | Read in the order the PDF stores its text, which is usually column by column. | Skim the imported text and fix any paragraph that reads out of order. |
| Password-protected PDFs | Not opened. Remove the password and import again. | Open it in your PDF app, save a copy without the password, import that. |
| More than 30 pages | The first 30 pages are read and the import tells you how many were skipped. | Split the check into sections anyway; that is more informative. |
| Your PDF | What happens | Account needed? |
|---|---|---|
| Exported from Word, Google Docs or a journal site | Read instantly in your browser | No, for checks up to 300 words |
| Longer than 300 words | Check up to 1,500 words at a time | Free account |
| Scanned or photographed pages | Text recognition reads each page | Free account, 10 pages a day |
| Password-protected | Will not open until the password is removed | n/a |
| Over 30 pages | The first 30 pages are read | Depends on length |
Frequently asked questions
Can I check a PDF for AI without uploading it?
Yes. The PDF is opened inside your browser and only its text is placed in the editor; the file itself never leaves your device. The text is sent for scoring only when you press Detect AI.
Is the PDF AI detector free?
Yes. The detector never spends words. Without an account it checks 300 words at a time; a free account raises that to 1,500 and unlocks reading scanned pages.
Can it check a scanned PDF?
Yes, with a free account. Pages that are images rather than text are read with OCR, 10 pages a day on a free account and 100 on paid plans. Check names and numbers in the recognized text before you score it.
Does converting a document to PDF hide AI writing from Turnitin?
No. A PDF stores the same words in its text layer, and Turnitin and other checkers read that layer. Changing the file type changes nothing about how the writing scores.
How long a PDF can I check?
The importer reads up to 30 pages and 20 MB. Each check covers 300 words without an account, 1,500 with a free one and up to 6,000 on paid plans, so long documents are checked in sections.
Why does the imported text show repeated headers?
Running headers and footers are part of each page's text, so they come in once per page. Page numbers are removed automatically, but titles and author lines are kept. Delete them before you check.
Is a high AI score proof that someone used AI?
No. Detectors estimate how closely text matches machine-written patterns, and formal or non-native writing is flagged more often than it should be. Use the score to decide where to look, then look at drafts and the writing itself.