Detect Hidden Text & Prompt Injection in PDF Files

No. The scan happens entirely locally in your browser, in a background worker. The file is never uploaded to a server.

Text embedded in a PDF’s content stream that’s aimed at an AI system reading the document — an instruction, a keyword, or a claim — rather than at the human looking at the rendered page. It works because most AI tools that "read" a PDF pull the same text layer this scanner checks, regardless of how that text is styled or positioned. It shows up most often in two places: resumes aimed at AI-based applicant screeners, and academic or reviewer documents aimed at AI graders or reviewing assistants.

Drop the file into the scanner above. It’s free, runs entirely in your browser, and checks all six known hiding techniques in one pass — there’s no sign-up, no page limit, and no cap on how many files you can scan.

Upload the file above and the report shows the extracted text for every finding directly, un-redacted, alongside the page number and which signal caught it. Most tools that offer to extract hidden text from a PDF online do it by uploading your file to a server first — this one never does; the extraction happens locally, in a background Web Worker.

A general term for text in a document that’s meant to be read by an AI system rather than a human — hidden via any of the same techniques covered here (invisible render mode, near-zero size, background-matched color, off-canvas position, zero-width Unicode, or a hidden layer). Prompt injection is the most common form of it in PDFs.

Yes, in two ways. Applicant-tracking systems and AI screeners increasingly watch for resume prompt injection and flag or reject a manipulated resume outright — the opposite of the intended effect. And any human reviewer who copies your resume text into another document, or runs it through their own tool, sees exactly what was hidden. If you want to check your resume for white text before sending it, use the scanner above.

Some educators intentionally plant a hidden instruction — a Trojan horse prompt — that’s invisible to a student but readable by an AI assistant, as a way to catch work generated by AI rather than written by the student, since the AI tends to follow the hidden instruction verbatim. This tool doesn’t build that trap for you; it’s built to detect hidden text in a PDF, which is exactly the technique you’d use to check whether a hidden prompt already exists in a document you’re reviewing, or to confirm one you planted took effect as expected.

It catches the structural techniques described above — invisible render mode, near-zero size, background-matched color, off-canvas position, zero-width Unicode, and hidden Optional Content Group layers. It doesn’t attempt to judge intent or interpret what any hidden text says, and a sufficiently unusual hiding technique outside these six signals could go undetected. Treat a clean result as "no known technique found," not an absolute guarantee.

Not necessarily. The tool reports structure, not intent. Some legitimate documents include intentionally hidden text — an invisible OCR text layer over a scanned image, for instance, or a legitimately tiny footer disclaimer. Look at what the extracted text actually says and where it came from before assuming the worst.

Text that is present in the PDF’s content stream — and readable by any tool that extracts text, including AI systems — while being invisible or illegible to a human looking at the rendered page. That’s different from a scanned page with no text layer at all, which this tool reports separately, since there’s nothing hidden there, just nothing extractable.

It checks the text layer, not rendered pixels. A scanned PDF with no embedded text at all has nothing to hide and nothing to scan — this tool reports that explicitly as "no extractable text," rather than a false "clean" result.