What a PDF prompt injection scanner can and cannot tell you
A hidden-text scan can identify visibility signals in a PDF. Here is what those findings mean, and what they cannot prove about intent or AI behaviour.
People use the phrase “PDF prompt injection” for a real concern: a document can contain text intended for an AI system rather than for the person reading the page. The phrase is useful, but it can make a simple check sound more certain than it is.
A scanner can show that text is hard to see, invisible, tiny, or placed beyond a page boundary. It cannot reliably tell you why that text is there, whether an AI service will follow it, or what a person meant when they created the PDF.
Visibility and intent are different questions
A white sentence on a white page is a visibility issue. A sentence asking an assistant to ignore its task is a content issue. They can occur together, but neither proves the other. A template mistake can leave harmless white text behind. A visible sentence can be inappropriate without being hidden at all.
That distinction matters when reviewing documents from students, vendors, clients, or public sources. Treating every unusual text object as an attack creates noise. Treating the visible page as the complete document creates blind spots.
What a useful scan can establish
A visual scan can give you concrete facts: this text box had almost no contrast with its surrounding page; this item used an invisible text mode; this font was unusually small; this object sat outside the visible page. Those facts are useful because they point you to a page and a region to inspect.
PDFShore's Hidden Instructions Scanner stays at that level. It renders the page in the browser, compares text regions with their local background, and shows a marked page thumbnail. It does not rank phrases, guess motivation, or label a document malicious.
How to review a flagged area
Open the original page and look at the nearby material. Does the text belong to a footer, a diagram, an accessibility layer, or an old edit? Is it relevant to the work you plan to do with the document? If the PDF came from outside your organization, make a clean copy before you send it to an AI system.
For a high-stakes decision, keep the evidence simple. Record the page number, save the source file, and ask the document owner about anything unexplained. A scanner is most helpful when it narrows the review, not when it pretends to replace it.
The practical rule
Check both layers of a PDF: what appears on the page and what the file exposes to extraction. Then make a decision from the document context. That is more dependable than trying to infer intent from a single line of text.