Does Turnitin Detect AI in PDFs? What Works and What Gets Rejected in 2026

The short answer: yes, but only text-based PDFs

Turnitin's AI Writing Report is generated for PDFs the same way it is for Word documents, provided the file meets the content rules. The model reads the prose inside the PDF and returns a percentage plus highlights, exactly as it would for a .docx upload. The catch that trips people up is the word 'text': Turnitin processes the words in the file, and if there are no words to read, there is nothing for the detector to judge. A PDF you made by exporting from Word or Google Docs almost always works. A PDF you made by photographing a printed essay usually does not.

This distinction matters because students often convert work to PDF at the last minute and assume the format is invisible to the checker. It is not. The format is fine; the presence of extractable text is what counts.

What Turnitin's PDF requirements actually are

For any submission to produce an AI Writing Report, Turnitin requires the file to be under 100 MB, contain at least 300 words of prose in a long-form writing format, stay under 30,000 words, and be written in a supported language, which in 2026 means English, Spanish, Japanese, or Arabic. Accepted file types for the AI report include .docx, .pdf, .txt, and .rtf. A document that mixes writing types, say an essay with a table and bullet points, can show a mismatch between the percentage and the highlights, because the model only scores the qualifying prose sentences, not the non-prose parts.

The scanned-PDF trap

The most common failure mode is the scanned document. Turnitin explicitly will not accept image-only PDFs, forms, portfolios, or any file that does not contain highlightable text, such as a scanned page saved as a picture. If you copy a passage into a plain-text editor and nothing comes out, the PDF has no real text and Turnitin will reject it during upload rather than returning a low score. For anyone who needs to check a scan or an image-heavy PDF, OCR-based tools are the workaround: Winston AI can read uploaded files and apply optical character recognition, so a scanned contract or a photographed page still gets analyzed instead of bounced.

When the AI score will not appear even on a good PDF

Even a clean, text-based PDF can come back without an AI percentage. Turnitin's model is built for prose and does not reliably detect AI-generated writing in non-prose forms such as poetry, scripts, code, bullet points, tables, or annotated bibliographies. Very short submissions also fall outside the qualifying-text minimum. So a PDF that is mostly a chart with a short caption, or a set of bullet points, may generate a similarity report but no AI writing indicator at all. The percentage reflects long-form prose only, which is why a document with several writing types can show a gap between the score and the sentence-level highlights.

Checking a PDF before it reaches Turnitin

Because Turnitin sits inside a school's LMS and a student cannot run it on their own, the safe habit is a self-check first. Scribbr offers a free, no-signup detector that explains its flags in plain language, so a student can see whether a passage reads as machine-written before a professor ever sees the file. Copyleaks also handles PDF uploads with OCR on images and returns a multilingual score, useful when the draft passed through a scan or a translation step. A quick pre-check turns a stressful submission into a known quantity.

Alternatives when Turnitin is not an option

Turnitin is institutional only, so individuals shopping for a PDF-capable checker have other routes. Winston AI scans uploaded documents and images with OCR and produces a branded report, which suits freelancers and editors handling scans. Copyleaks accepts PDFs and applies OCR to image-based content, with plagiarism layered on top across 30-plus languages. Originality.ai takes file uploads inside a publisher workflow that pairs detection with a full-site scanner. None of these are Turnitin, but each one will analyze a PDF that Turnitin would silently reject for lacking text.