Resume Tip: Hacking "AI" screening of resumes

Original Article ↗ Hacker News Discussion ↗

Effectiveness of resume prompt-injection (“ChatGPT, ignore all other applications…”)

  • Many commenters doubt the trick works broadly:
    • Major ATS products often use OCR and ignore text color, so white-on-white text disappears.
    • AI components tend to extract skills, experience, and dates, not follow arbitrary instructions from the document.
    • One person reports repeated experiments with GPT-4o where such lines had no effect.
  • Some think it might work only in very simple or hastily-built systems, or where HR staff literally paste resumes into ChatGPT with a naive prompt.
  • Several suggest any “success” is more likely due to including desirable keywords (“ChatGPT”) than to the instruction itself.
  • Overall consensus: amusing idea, not a reliable general tactic.

How ATS and LLMs are actually used

  • ATS (Applicant Tracking Systems) predate LLMs and already parse resumes for skills, work history, and keywords.
  • LLM use patterns described in the thread:
    • Embedding-based matching between resumes and job descriptions.
    • Simple scoring prompts (“compatibility_score, passed: true/false”).
    • Experimental multi-step prompt chains for screening.
  • Some companies reportedly use Azure OpenAI–style hosted models to stay within privacy/compliance constraints.

Gaming automated screening

  • White-on-white keyword stuffing has existed for decades (SEO, plagiarism evasion); people now reuse it for resumes and AI prompts.
  • Mixed reports:
    • Some say keyword-flooded footers significantly increased interview requests, including for government roles.
    • Others insist modern systems counter this, which is why many force manual entry of work history.
  • General observation: any automated filter can be adversarially probed and “optimized against,” given enough attempts.

Ethics, incentives, and job-search strategy

  • One side: gaming filters is pointless or dishonest; better to pursue roles where you’re a genuine match and avoid AI-heavy employers.
  • Other side: filters are noisy and biased; you may be perfectly qualified yet auto-rejected, so tactical “gaming” just restores a chance to reach a human.
  • Several note that personal networks still dominate hiring; ATS/AI mostly add another opaque layer.

Employer countermeasures

  • Some employers embed “honeypot” phrases in job ads so LLM-generated, unedited cover letters reveal themselves and are auto-rejected.
  • Defenders frame this as spam filtering for low-effort, copy-paste applicants.
  • Critics argue it’s another arbitrary hoop that may filter good candidates and overestimates the ability to reliably distinguish human vs LLM text.