Do AI detectors work on scanned PDFs? The 18 cards that accept a PDF upload, the four that say the file needs a text layer, and the three that OCR images instead
None of the 72 AI detectors in the index states that a scanned PDF, a file whose pages are pictures of text with no text layer, is read by optical character recognition and scored, by the input limits and file types fields of every card read 10 November 2026 in the data file. Eighteen cards accept a PDF upload, and four of the eighteen say what happens to a scan: Sapling takes a "PDF or Word document with selectable text" (sapling.ai, read 8 November 2026), ZeroGPT takes a "text-based .pdf" (zerogpt.com, read 8 November 2026), Detecting-AI.com states that "scanned PDFs need OCR and are not supported yet" (detecting-ai.com, read 10 November 2026), and Turnitin's FAQ says that "for AI detection we're focusing on typed text", leaving handwritten work to Gradescope, which has no AI detection (guides.turnitin.com, read 10 November 2026). Three other cards state OCR for images or handwriting and name no PDF at all: Winston AI, Compilatio and Copyleaks.
A PDF is a container, and the question is what is inside it. A PDF exported from a word processor carries a text layer that a detector can read; a PDF made by a scanner or a phone camera carries page images, and a text detector sees nothing until something runs OCR. The first table lists every card whose input limits or file types field names PDF as an upload, quotes the field, and records in the third column what the card says about scanned pages, which for 14 of the 18 is nothing. The second table lists the cards that state an OCR or handwriting route and no PDF, because for a scan they are the only stated path. This page is one of the questions answered from the index; the file types ranking counts every format each card states, the input limits ranking carries the size and word ceilings, and the AI detector register carries every card.
| Detector | PDF upload as the card states it | What the card states about scanned pages or OCR | Source | Read date |
|---|---|---|---|---|
| Pangram | "Supports PDF, DOCX, RTF files and up to 100 files at a time" | Not stated | pangram.com | 8 November 2026 |
| Sapling | "Files: PDF or Word document with selectable text"; API file endpoint takes .pdf up to 30 MB | Selectable text required: a page image with no text layer is outside the stated input | sapling.ai | 8 November 2026 |
| JustDone | "Paste text directly or upload a PDF, Word, or TXT file" | Not stated | justdone.com | 8 November 2026 |
| Originality.ai | "Paste, upload (docx, pdf, doc) or URL" | Not stated | originality.ai | 8 November 2026 |
| ZeroGPT | "Paste text or upload .txt, .docx, text-based .pdf" | "text-based .pdf": the page limits PDF input to files with a text layer | zerogpt.com | 8 November 2026 |
| Proofademic | "Paste text or upload .docx, .txt or PDF; Batch Scan PDF, DOCX, TXT" | Not stated | proofademic.ai | 8 November 2026 |
| Turnitin | "accepted file types .docx, .pdf, .txt, .rtf"; file size under 100 MB; at least 300 words of prose | FAQ: "for AI detection we're focusing on typed text"; handwritten text is Gradescope's use case and Gradescope gets no AI detection | guides.turnitin.com | 10 November 2026 |
| AIDetectPlus | "upload .docx, .pdf, .txt and .html, with separate detectors for essays, papers, blog and SEO copy, resumes and PDFs" | Not stated | aidetectplus.com | 9 November 2026 |
| Smodin | "paste text or upload .txt, .doc, .docx or .pdf" | Not stated | smodin.io | 9 November 2026 |
| Quetext | "bulk uploads of 20 or 100 files in .doc, .docx, .pdf or .txt" | Not stated | quetext.com | 9 November 2026 |
| TruthScan | "one result is one image or one PDF page" on the image and document detector; "Word, PDF and Google Docs document processing for enterprise users" on the text detector | Not stated as OCR: a PDF page is scored as an image for forgery and AI generation, which is a different question from reading its words | truthscan.com | 9 November 2026 |
| Decopy | "upload a file in .txt, .pptx, .ppt, .doc, .docx or .pdf" | Not stated | decopy.ai | 9 November 2026 |
| GPTKit | Upload in 19 types including .pdf, .jpeg, .png and .gif | Not stated: image files are accepted and no OCR step is described | gptkit.ai | 9 November 2026 |
| PlagiarismCheck | API: "plain text or a file in doc, docx, pdf, txt, odt or rtf" | Not stated | plagiarismcheck.org | 9 November 2026 |
| Isgen | "Bulk Scan accepts up to 150 files in pdf, docx or doc at 10 MB each" | Not stated | isgen.ai | 9 November 2026 |
| Fast-DetectGPT | Hosted demo: "a .txt, .pdf or .docx file of up to 10 MB" | Not stated | fastdetect.net | 10 November 2026 |
| Detecting-AI.com | "upload TXT, PDF or DOCX"; "PDF and DOCX text is read privately in your browser" | "scanned PDFs need OCR and are not supported yet" | detecting-ai.com | 10 November 2026 |
| Trinka | "Upload your MS Word or PDF file to check for AI content"; marketplace files up to 100 MB and 25,000 words | Not stated | trinka.ai | 10 November 2026 |
Inputs as the input limits and file types fields of each card record them from the page linked, read on the date shown; the Turnitin row quotes the FAQ re-read on 10 November 2026, and the index has not uploaded a scanned file to any of the eighteen.
Which AI detectors accept a PDF upload at all?
Eighteen of 72, counted from the two fields above on 10 November 2026, and the ceilings differ by an order of magnitude. Turnitin takes files under 100 MB with at least 300 words of prose and at most 30,000 words (guides.turnitin.com, read 10 November 2026); Pangram takes up to 100 files at a time (pangram.com, read 8 November 2026); Isgen takes 150 files at 10 MB each in a bulk scan (isgen.ai, read 9 November 2026); Sapling's API file endpoint takes a file up to 30 MB (sapling.ai, read 8 November 2026); Fast-DetectGPT's hosted demo takes 10 MB (fastdetect.net, read 10 November 2026). The bulk and batch view reads the file counts, and the file types ranking sorts the cards by how many formats they state, where Sensity's 22 media formats lead and the text cards with a PDF route carry between 2 and 10.
Fourteen of the eighteen say nothing about what is inside the PDF. Silence is not acceptance: a detector that reads the text layer of an exported PDF will receive an empty document, or a few words from a header, when the pages are images, and the score it returns describes that empty input. Only Detecting-AI.com's page says so in plain words, and only Sapling's and ZeroGPT's pages restrict the input to selectable or text-based files, which is the same statement made as a requirement (detecting-ai.com, read 10 November 2026; sapling.ai, zerogpt.com, read 8 November 2026).
Which detectors read images of text or handwriting?
Three cards state an OCR route, and none of the three names PDF as an input. Winston AI lists "handwriting (OCR)" among its modalities, takes uploads in .docx, .png and .jpg, and includes OCR in the Essential plan at $18 a month list price (gowinston.ai, read 8 November 2026). Compilatio states OCR on handwritten documents inside Magister, the institutional product, with no file list published (compilatio.net, read 9 November 2026). Copyleaks lists "OCR files: gif, png, bmp, jpg, jpeg" beside its image formats and no text document format on the detector page (copyleaks.com, read 8 November 2026). For a scanned PDF, these three are the stated path, and it runs through a page image rather than the PDF: export each page as a .png or .jpg and upload the image, which Winston AI and Copyleaks state and Compilatio's page does not detail. Turnitin's answer is the opposite: its FAQ says the detection "is focusing on typed text", and the sister product that reads handwriting, Gradescope, has no AI writing detection on any assignment type, as the Gradescope question sets out (guides.turnitin.com, read 10 November 2026).
Cards that state OCR or handwriting input and name no PDF
| Detector | OCR or handwriting as the card states it | Inputs stated | Source | Read date |
|---|---|---|---|---|
| Winston AI | Modalities: text, image, "handwriting (OCR)"; Essential plan includes "OCR, PDF reports" | Uploads in .docx, .png and .jpg; up to 200,000 characters a scan on paid plans | gowinston.ai | 8 November 2026 |
| Compilatio | "handwriting (OCR on handwritten documents, Magister)" | Documents uploaded to Magister+ by teachers or Studium by students; no file list published | compilatio.net | 9 November 2026 |
| Copyleaks | "OCR files: gif, png, bmp, jpg, jpeg" | Image files JPG, JPEG, PNG, WEBP, BMP, HEIC, HEIF; web platform minimum 255 characters, maximum 2,000 pages | copyleaks.com | 8 November 2026 |
Routes as the modalities, plans and input limits fields of each card record them, read on the date shown. The other 51 cards name neither a PDF upload nor an OCR route: 18 + 3 + 51 = 72. Most of the 51 are paste-only text tools such as GLTR and Grammica, research code such as DetectGPT and Binoculars that takes text from a script, or media detectors whose inputs are images, audio and video, which the modalities view lists.
Does a scanned PDF change the score?
No study in the index has measured it. The 35 records in the studies index scored pasted or exported text, images, audio and video; none fed a detector a scanned document, and no card quotes a vendor test on OCR output. What the records do show is that input length and form move scores: Turnitin excludes "lists, headers, references, tables, formulas and code" from scoring and needs 300 words of prose (guides.turnitin.com, read 10 November 2026), Sapling flags results on text under 500 characters as low confidence (sapling.ai, read 8 November 2026), and the bullet points and code question lists the ten cards that state how non-prose input is treated. OCR errors, broken line endings and column order are the same kind of non-prose noise, with no vendor statement on their effect.
A scanned PDF reaches a text detector only through an OCR step that no card in the index states it performs, so a score returned on such a file describes whatever text the detector found, which may be none.
Questions about scanned PDFs and AI detectors
Can an AI detector read a scanned PDF?
0 of 72 cards state that a scanned PDF is read by OCR and scored, by the input limits and file types fields read 10 November 2026; 18 accept a PDF upload, and the four that address scans, Sapling, ZeroGPT, Detecting-AI.com and Turnitin, require a text layer or typed text (sapling.ai, zerogpt.com, detecting-ai.com, guides.turnitin.com).
Which AI detectors accept PDF files?
18 of 72: Pangram, Sapling, JustDone, Originality.ai, ZeroGPT, Proofademic, Turnitin, AIDetectPlus, Smodin, Quetext, TruthScan, Decopy, GPTKit, PlagiarismCheck, Isgen, Fast-DetectGPT, Detecting-AI.com and Trinka, by the fields read 10 November 2026 in the data file. The handwriting and OCR view, added 10 November 2026, lists the three cards that state OCR of handwriting or image files and the eight that take an image without reading it. The file types ranking sorts them by the number of formats each states.
Which detectors do OCR on handwriting or images of text?
3 of 72: Winston AI ("handwriting (OCR)", uploads in .docx, .png and .jpg), Compilatio (OCR on handwritten documents in Magister) and Copyleaks ("OCR files: gif, png, bmp, jpg, jpeg"), none of which names PDF (gowinston.ai, compilatio.net, copyleaks.com, read 8 and 9 November 2026).
Does Turnitin check handwritten or scanned work for AI?
No. Turnitin's FAQ says that "for AI detection we're focusing on typed text" and that Gradescope, whose "primary use case ... is handwritten text", will not get the capability (guides.turnitin.com, read 10 November 2026); the accepted types are .docx, .pdf, .txt and .rtf with at least 300 words of prose.
Has any study tested a detector on scanned documents?
None of the 35 records in the index did: every record scored pasted or exported text, images, audio or video, by the study pages read 10 November 2026. The nearest evidence is on non-prose input, which the bullet points and code question collects.
Sources
All read 10 November 2026 unless an earlier date is shown; the source of every row is linked in the tables.
- Pangram home page (file inputs): https://www.pangram.com/ (read 8 November 2026)
- Sapling AI detector (file inputs): https://sapling.ai/ai-content-detector (read 8 November 2026)
- JustDone AI detector: https://justdone.com/ai-detector (read 8 November 2026)
- Originality.ai pricing (file inputs): https://originality.ai/pricing (read 8 November 2026)
- ZeroGPT pricing (file inputs): https://www.zerogpt.com/pricing (read 8 November 2026)
- Proofademic AI text detector: https://proofademic.ai/ai-text-detector/ (read 8 November 2026)
- Turnitin AI writing detection FAQs: https://guides.turnitin.com/hc/en-us/articles/28477544839821-Turnitin-s-AI-writing-detection-capabilities-FAQs
- AIDetectPlus home page: https://aidetectplus.com/ (read 9 November 2026)
- Smodin AI content detector: https://smodin.io/ai-content-detector (read 9 November 2026)
- Quetext integrations (file types): https://www.quetext.com/integrations (read 9 November 2026)
- TruthScan text detection API documentation: https://truthscan.com/truthscan-ai-text-detection-api-documentation (read 9 November 2026)
- Decopy AI detector: https://decopy.ai/ai-detector/ (read 9 November 2026)
- GPTKit home page: https://gptkit.ai/ (read 9 November 2026)
- PlagiarismCheck AI detector: https://plagiarismcheck.org/ai-detector/ (read 9 November 2026)
- Isgen pricing (bulk scan): https://isgen.ai/pricing (read 9 November 2026)
- Fast-DetectGPT hosted demo: https://fastdetect.net/
- Detecting-AI.com home page: https://detecting-ai.com/
- Trinka AI content detector: https://www.trinka.ai/features/ai-content-detector
- Winston AI pricing (uploads, OCR): https://gowinston.ai/pricing/ (read 8 November 2026)
- Compilatio Magister: https://www.compilatio.net/en/magister (read 9 November 2026)
- Copyleaks AI content detector (OCR files): https://copyleaks.com/ai-content-detector (read 8 November 2026)
Revision log: 2026-11-10: page created from the input limits, file types and modalities fields of 72 cards at index version 2026.11.12, with the Turnitin FAQ re-read on 10 November 2026; 2026-11-10: handwriting and OCR view linked
Related pages: Which AI music detectors exist? The five cards that state music among the media they score, the generators each names, and the zero study records behind them, Which AI detectors are free for nonprofits or education? The 12 cards that sell an institutional programme, the one university platform that charges nothing, and the two nonprofit tools that closed, AI detectors ranked by the upload formats the vendor states and Questions about AI detectors, answered from the index.