Can you tell if an AI tool made a PDF?
Sometimes, from clues in the file. This check reads the hidden "made with" details a PDF stores and tells you if they name software that AI tools often use to build PDFs. It never reads the writing, and ordinary invoices and reports can be flagged, so a result is a clue and not proof of who wrote anything.
How do I check if a PDF was made by ChatGPT or another AI tool?
Upload the PDF. You get a score from 0 to 100, a level from No Indicators to Strong Indicators, the software the file says it was made with, and a plain list of the reasons behind the score.
It is free and works without an account. The free plan allows 100 uploads per calendar month after email verification and files up to 50 MB.
What does the AI detection look at?
Every PDF records which program it was written in and which program saved it. The report shows these as Creator and Producer. The check reads those two details and compares them with a list of software that AI assistants commonly use when they build a PDF for you.
When ChatGPT, Claude or a similar assistant produces a PDF, it usually does so by running code, and the code library leaves its name in the file. ReportLab, WeasyPrint, pdf-lib, PDFKit, Puppeteer and Playwright are typical examples. The check also notices the name of an AI product itself in those details, and it rates a file higher when a match comes with an empty author or title, because generated files often leave both blank.
What does a high or low result mean?
The score counts clues. It is not the chance that a person or an AI wrote the document.
- No Indicators: the "made with" details name nothing on the list. Word, LibreOffice and most desktop apps land here.
- Few Indicators: a weak clue, such as software that is occasionally used by scripts.
- Some Indicators: the file names software that automated and AI workflows often use. From this level up, the report says AI-generator indicators were found.
- Strong Indicators: the file names software closely tied to AI-built PDFs, often with no author or title filled in.
Can a normal PDF be flagged as AI-made?
Yes, and it happens often. The same software is used by plenty of systems that have nothing to do with AI. Invoicing platforms build bills with headless Chrome, reporting servers produce statements with ReportLab or WeasyPrint, and web apps save forms and exports with pdf-lib. An invoice from your accounting system or a report downloaded from a dashboard can show Some or Strong Indicators.
So a flagged file tells you it was produced by automated software. It does not tell you a language model wrote the words.
Can an AI-written PDF show no indicators?
Yes. If someone copies text from an AI assistant into Word or Google Docs and exports it the usual way, the file names the word processor and the result is No Indicators. The "made with" details can also be removed or retyped in seconds with any metadata editor.
No Indicators means the file names no software on the list. It says nothing about who wrote the content, and it is not evidence that a person did.
Does it read the text or check the writing style?
No. The words on the pages, the fonts, the images and the writing style play no part in the result. This is not an AI text detector, and it cannot tell you whether an essay, a cover letter or a report was written by a person. If the writing is what you need to judge, extract the text and read it yourself.
What should I do with a flagged PDF?
Treat it as a reason to ask how the document was produced, not as an accusation. A student, supplier or applicant may have used an online tool that saves PDFs with the same software an AI assistant uses.
- Ask the sender which program or service they used, and whether that matches the software named in the report.
- Open the full metadata report to see the author, the dates and the other details stored in the file.
- Ask for an earlier draft or the original document, then compare the two files.
- Do not rely on this result alone for a decision about grades, hiring or payment.
Last updated . Written from the detector's source code and from test PDFs we built for the purpose.