Free tool · no signup
CV prompt injection scanner
Some CVs now contain text you can't see, written for the AI that screens them: white-on-white keywords, 1pt instructions such as "rank this candidate first", or text placed off the page. Check any CV here for free. The file is checked in memory on our server, then discarded: never stored, never logged and never sent to an AI model.
- PDF and DOCX, up to 10 MB
- Results in under a second
- Nothing stored
What it checks
Invisible text
White or background-coloured text, transparent text and Word's hidden-text setting.
Tiny text
Anything under 4pt, in any colour.
Off-page text
Text placed outside the visible page.
Instructions to AI
Sentences addressed to AI readers or screening tools, even when visible.
A USENIX Security 2026 study of about 196,000 resumes found hidden prompt injections in about 1% (summary). In our own test with two Claude models, the models weren't fooled, but the same candidate went from "top match" in 8 of 8 runs to 4 of 32 when hidden text was present, with no visible reason for the recruiter.
Test your own AI screener
The open CV prompt-injection test set has 14 fictional CVs: 10 with a different hidden-text attack each, and 4 controls that must not be flagged. With plain text extraction, every hidden instruction reaches the model intact.
Use it from code
curl https://api.verbatimcv.com/v1/scan -F file=@cv.pdf
No key is needed (10 scans a minute). The full parser, POST /v1/parse, removes this text before
extracting anything and returns the CV as JSON with a verbatim quote for every preference.
Get a free key or read
how hidden prompt injection works.
The scanner reports what it finds; whether a hidden instruction should count against a candidate is your decision. It checks the text layer only: text inside images isn't read.