Does hiding prompts or white text in a resume work?

Not reliably, and it is easy to catch: text extraction ignores colour, so hidden lines show up in the parsed profile and any plain-text view. Screening vendors scan for hidden text, and some recruiters reject applicants who use it.1

Written 8 October 2026. CitedCV. How we write our guides

Check your CV for ATS problems, no account needed

Drop a PDF, DOCX or text file, or paste the text. You get a score, the problems a parser trips on, and your CV as a parser files it.

Runs in your browser. The file and its text never leave this device.

What resume prompt injection is

Resume prompt injection is text hidden in a CV that a person is not meant to see but software reads: text coloured to match the background, a font too small to notice, or content tucked into the file's data. A May 2026 study by researchers from the University of North Carolina, Duke University, hireEZ and others, based on about 200,000 real resumes from the recruiting platform hireEZ, sorts it into two kinds:

Instruction injection: a sentence addressed to an AI model, telling it how to rate the applicant.
Data injection: hidden content aimed at keyword matching, most often skill lists, but also copied job requirements, degrees, certifications and experience descriptions. More than 90% of the injections the study detected were of this kind.

OWASP, the open security project, lists prompt injection as LLM01, the first entry in its 2025 Top 10 for large language model applications. It defines it as input that alters a model's behaviour or output in unintended ways, and notes that such input can affect the model even when it is imperceptible to humans. When the input arrives in a file, OWASP calls it indirect prompt injection; one of its example attack scenarios is a resume carrying split prompts meant to win a positive recommendation.

Why people try it

Because many employers use AI to sort applications, and a hidden line costs nothing to add. The New York Times reported in October 2025 that the tactic was being shared in TikTok videos and Reddit forums, and that it builds on an older habit of hiding keywords such as 'communication' or 'Microsoft Excel' in invisible text.

What to do in the first minute

Check the file you send for text you cannot see, even if you never added any: Duke's summary of the study warns that applicants might not know a template they use contains prompt injection commands. Drop your CV into the free checker at the top of this page. It runs in your browser with no account, and its parser view lists every line it extracted; for a PDF, that includes text coloured white or set in a tiny font.

Drop the PDF or DOCX you plan to send into the checker, or open the file and select all of its text.
Read the parser view, or paste the selection into a plain text editor, from top to bottom.
Compare it with what you see on the page. A line you cannot see on the page is hidden text.
Delete it in the source document, export again, and check the new file once more.

Why hidden text shows up anyway

Text extraction reads the characters in a file, not how they look. The study's own examples show it: an instruction set in a 1-point font at the bottom of a page was imperceptible to human readers but extractable by PDF text parsers, and skill keywords coloured to match the background were invisible on the page but present in the machine-extracted text. The hidden line lands wherever the visible text lands.

Where a recruiter can see it

In the candidate profile an applicant tracking system fills from the parsed text, and in any plain-text copy of the resume.
In the file itself, once someone changes the colour: The New York Times described a recruiter in Britain who spotted a white-text instruction only because he had set the resume's font to black for review.
In a risk flag, where the employer runs a tool that checks for hidden content (see the defences below).

What CitedCV's checker does with it

The free checker reads a PDF with pdf.js and keeps every text item pdf.js returns; the code applies no colour or size filter. A DOCX is read as raw text. Zero-width and control characters are stripped before the checks run, and every remaining non-empty line appears in the parser view under a section. The checker shows hidden text but does not label it as hidden, so compare the parser view with your page.

Why it is risky

Vendors scan for it. Greenhouse's help page for the Nova by Dweet integration (legacy version, updated 27 August 2026) lists hidden content detection, covering invisible text in tiny font sizes, keyword stuffing and overlapping duplicate text layers, among its resume fraud checks; results go back to Greenhouse as a note and a risk tag.
Detection is in production. The study's authors say both of their detectors, one of which compares what a person sees with what a machine extracts, have been integrated into hireEZ's production systems.
Recruiters reject it. The New York Times quoted a recruiter at the e-commerce company Commercetools who rejects candidates when she finds hidden text, and reported that some recruiters automatically reject applicants who try to trick their AI.
It reads as dishonesty. A hidden skill, degree or job description claims something you did not show the reader. If it is not true, it is a false statement on an application; if it is true, it belongs in visible text.
The target moves. Duke's summary says earlier AI models were especially vulnerable because they often treated hidden messages as legitimate instructions, and OWASP lists defences that application builders use against it.

Does it ever work?

Some people say it did. The same New York Times report quotes job seekers who said they got more interviews after adding hidden prompts, and one who said at least one recruiter rejected him after finding it. These are anecdotes without a comparison. The hireEZ study, the first large-scale look at real resumes, found hidden injections in about 1% of them, a lower-bound estimate, and a share that rose noticeably over the past one to two years. It did not measure whether the injections changed screening outcomes.

How screening tools defend against it

Defence	Documented by	What it does
Hidden content detection: tiny fonts, keyword stuffing, duplicate text layers	Greenhouse help page for Nova by Dweet (legacy)	Adds a note and a risk tag to the candidate in Greenhouse
Comparing what a person sees with what the machine extracts	The hireEZ study (UNC, Duke and others)	Flags text present in the extracted copy but not on the visible page
Separating untrusted content, constraining the model, human approval for high-risk actions	OWASP LLM01:2025	Limits what an injected line can make a model do
Flagging lines with injection patterns such as "ignore previous instructions" or a "system:" role marker; stripping zero-width characters	CitedCV code (intake and upload checks)	No fact may come from a flagged line

The study adds a caution for defenders: because most injections are hidden data rather than instructions, detectors that look only for explicit instruction patterns are likely to miss most of them. CitedCV's pattern check is that kind of check. Its other guard is that every fact read from an uploaded CV is shown to the person to confirm, so hidden keywords in your own file come back to you, not quietly into the CV.

What to do instead

Use the posting's own terms in visible text where they are true for you: the same skill names, tools and titles.
State evidence: what you did, with what, and a number where you have one. That is what a recruiter reads for, and a matching tool can find it.
Keep the file clean: one column, contact details in the body, standard headings, and no text you would not want a recruiter to read.
Explain a career change or a gap in a cover letter rather than hiding keywords.
If you think an AI will misjudge you, ask for human review where the employer offers it; the page on opting out of AI resume screening explains when that is possible.

Where CitedCV fits

If you rebuild, CitedCV writes a CV only from facts you give it: each line carries the ids of the facts it came from, a deterministic verifier checks numbers and names against those facts, and lines it cannot verify are dropped. When you build for a posting, the review page lists the posting's keywords missing from your CV, so you can add the ones that are true in plain sight. Tailoring by hand works too.

Who this page is not for

You want instructions for hiding text in a resume. This page does not give them.
You build or buy AI screening and need defences. Start with OWASP's LLM01:2025 entry and the hireEZ study's detection methods.
You want to know whether your CV parses at all. That is a layout question; see the page on resumes that do not parse.

Alternatives to hidden text

Each of these does in the open what hidden text tries to do in secret.

Option	What it does	Limit
Tailoring by hand	Puts the posting's true terms in visible text	Takes time for each application
A cover letter	Explains context a keyword list cannot	Not every form has a field for it
Asking for human review	Removes the AI score where the employer offers it	Many employers do not offer it
Checking the file before sending	The free checker at the top lists every extracted line	It does not run any employer's parser or AI
A CV builder that writes from your facts	CitedCV is one; the comparison page names others	Each tool reads your file its own way; check what it keeps

Questions people ask

Does white text on a resume work?

Not reliably. White text is extracted like any other text, so it appears in the parsed profile and any plain-text copy, and some recruiters reject applicants when they find it.

Can ATS detect hidden text in a resume?

Yes, some tools can. Greenhouse documents a Nova by Dweet integration that detects tiny-font text, keyword stuffing and duplicate text layers, and the hireEZ study's detectors run in hireEZ's production systems.

What is resume prompt injection?

It is text hidden in a resume to steer an AI screener: an instruction addressed to the model, or invisible data such as skill lists. OWASP lists prompt injection first in its Top 10 for LLM applications.

Does "ignore previous instructions" work on AI resume screeners?

There is no evidence it works reliably. Duke's summary says earlier models were especially vulnerable, OWASP documents defences against it, and the large hireEZ study did not measure whether injections changed outcomes.

Will I be rejected for hidden text on my resume?

You can be. The New York Times quoted a recruiter who rejects candidates when she finds hidden text, and the Nova integration for Greenhouse adds a risk tag when it detects hidden content.

How do I check my resume for hidden text?

Compare what you see with what software extracts. Paste the whole file into a plain text editor, or use the free checker at the top of this page, whose parser view lists every extracted line.

Can a resume template contain hidden prompts?

Yes, it can. Duke's summary of the study warns that applicants might not know a template they use contains prompt injection commands, so check any downloaded template before you send it.

Does CitedCV remove hidden text from my CV?

Partly. It strips zero-width characters and flags lines with injection patterns so no fact comes from them; white text is extracted like any text and comes back to you as facts to confirm.

Sources

OWASP GenAI Security Project, LLM01:2025 Prompt Injectiongenai.owasp.org/llmrisk/llm01-prompt-injection/
Zhang, Jia, Tan, Jiang, Gong, Chen and Song, Measuring Real-World Prompt Injection Attacks in LLM-based Resume Screening (arXiv, May 2026)arxiv.org/html/2605.28999v1
Duke Today, Tricking AI in the Job Hunt (10 August 2026)today.duke.edu/2026/08/tricking-ai-job-hunt
Greenhouse Support, Nova by Dweet integration (legacy)support.greenhouse.io/hc/en-us/articles/46531208072219-Nova-by-Dweet-integration-legacy
The New York Times, Recruiters use AI to scan resumes. Applicants are trying to trick it. (October 2025, republished by The Star)www.thestar.com.my/tech/tech-news/2025/10/09/recruiters-use-ai-to-scan-resumes-applicants-are-trying-to-trick-it