How to detect white text and hidden AI prompts in résumés
Copy all the text out of the résumé into a plain text editor and compare it with the page: white text and tiny text show up in the copy.
That takes a minute for one résumé. For hundreds a day you need a check that runs before your screening software reads them, because most hidden prompts do not look like instructions at all.
The free way to check
For a PDF or a Word résumé:
- Press Ctrl+A (Cmd+A on a Mac), copy, and paste into Notepad or TextEdit in plain text mode. Compare the paste with the page.
- In a Word résumé, select all and set the font colour to Automatic and the size to 12. Hidden text appears on the page.
- Watch for any block of text you did not see on the page, whether or not it reads like an instruction.
In a study of 196,682 real résumés submitted to a live hiring platform, roughly 1% carried a hidden prompt — white-on-white text, 1-point fonts or text placed off the page — and more than 90% of those prompts contained no explicit instruction. A search for “ignore previous instructions” misses most of them.
Where the free way falls short
- Nothing in the copy marks which lines were invisible. You have to compare the whole résumé, line by line.
- Text behind a photo or a coloured box, and text placed off the page, stay out of sight in the document itself.
- Invisible Unicode characters paste as nothing, so the copy looks clean.
- Nobody compares hundreds of résumés by hand every day, and screening software reads every one.
How HiddenContent.ai checks it
HiddenContent.ai checks each résumé file — PDF or Word — before your screening software reads it. It reports the hidden text itself, where it sits and the trick behind it, so a recruiter can read it and decide.
It marks hidden text written as an instruction to a model, but it reports all hidden text, because most of it is not written that way. It also returns a clean copy of the text, with the hidden parts removed, for your screening software to read instead.
No file leaves our machines, and no AI model reads your document. See one real analysis, end to end for exactly what comes back. Sign up with GitHub for 20 free pages, then a cent a page.
Real cases
- Measuring Real-World Prompt Injection Attacks in LLM-based Resume Screening — USENIX Security 2026, 27 May 2026
- RAPIDS: Resume Attack Prompt Injection Detection at Scale — ACL 2026 Industry Track, 2 Jul 2026
- Job applicants are hiding secret AI messages in their résumés — Business Insider, 6 Sept 2026
The techniques, one page each
- Text the colour of its background
pdf.white-text - Type too small to read
pdf.tiny-font - Text placed outside the page
pdf.offpage-text - Text covered by something opaque
pdf.occluded-text - Text that matches its background
docx.white-text - Text below readable size
docx.tiny-font - Zero-width characters in text
unicode.zero-width - Hidden text in hiring: every case we have found
Questions
How common are hidden prompts in résumés?
In a study of 196,682 real résumés, roughly 1% carried one.
Is white text in a résumé always an attack?
No. Light text on a dark banner is a design choice, and some hidden text is harmless. What matters is what the text says and whether the person reading the résumé could see it, which is why HiddenContent.ai shows you the text itself rather than a score.
Do applicants really do this?
Yes. One applicant hid about 1,500 characters of white text telling screening software to declare them a top-tier fit: see In the Wild.