Find what's hiding in your PDF

Text colored so close to the page background that a human reader sees nothing, while an AI parsing the text layer reads it in full. White text on a white page is almost never legitimate.

Text drawn with PDF render mode 3, which paints no pixels at all yet stays in the machine-readable text layer. Normal authoring tools essentially never use this.

Text set below 4pt — too small to read on screen or in print, but fully extractable by software. Occasionally a legitimate figure callout dips this small, so read the text before judging.

Content hidden by cropping is surfaced through the crop box mismatch check below. Text that sits fully outside the visible page area is clipped by PDF engines — this one included — before it ever reaches text extraction, so it cannot be quoted directly here.

The page is cropped smaller than its full media size, which can push content into the trimmed margins. Frequently benign in scanned files — inspect what falls outside the crop.

Optional-content layers that are off by default or explicitly marked hidden. The layer contents stay in the file and remain readable by software even though they are not shown.

One or more files attached inside the PDF. Routine on engineering datasheets, unusual on papers or contracts. The names are listed below — do not open them without checking.

The document carries JavaScript, which can run when the file is opened in some viewers. Expected in interactive forms, out of place in a static paper or contract. It is never executed here.

Comments and sticky notes whose text is invisible in normal rendering but still sits in the file. Often legitimate, but a classic place to hide instructions — always read the content.

Text matching known patterns that try to steer an AI reviewer or summarizer — for example, telling it to ignore its instructions or return a positive verdict. Suggestive, not conclusive: a paper about prompt injection may quote such strings legitimately, so read the surrounding context.