If you have asked a chatbot to polish your resume, you may have handed part of your career story to a system that holds assumptions about who you are. A study published in PNAS, as reported by Phys, finds that major AI models can reproduce gender stereotypes both when they generate resumes and when they assess candidates. By the end of this article, you will know where the bias tends to hide and what to check before trusting AI with your job search or your hiring pipeline.
What the research found
According to Phys's report on the study, the problem can start before an application is even submitted. A woman asks an AI system to help write her resume. The system chooses words, highlights qualities and builds a profile. That is where stereotypes can slip in.
The research also points to bias during hiring itself, when AI is used to assess candidates. In short, the same kind of skew can show up on both sides of the process: the candidate's side (how you are described) and the employer's side (how you are judged).
The full details of the models tested and the size of the effects are beyond what we can report here, so check the PNAS paper if you need specifics.
Why "fair on average" can still be unfair to you
This is the most useful idea in the study. Per the reporting, outcomes that look balanced on average can hide penalties in gender-skewed professions. Women entering male-dominated fields are especially exposed.
Here is a plain way to think about it. Imagine a tool that treats men and women equally when you look across all jobs combined. That headline number can still mask what happens in one specific field, such as a role where most workers are men. A woman applying there may be penalized even though the overall score looks fine.
If you are job hunting in a field that is dominated by one gender, an "it tested fair overall" claim tells you very little about your case.
Check your AI-written resume before you send it
The research does not give a checklist, so treat the steps below as practical suggestions drawn from what it found, not as findings from the study.
Read for stereotyped language
Look at the qualities the AI chose to highlight. Does it emphasize warmth, collaboration and support when you would rather lead with technical skill, results or leadership? If the tone does not match the role you want, rewrite it.
Compare against your own facts
Go line by line and ask whether each claim comes from your actual experience. Keep achievements, numbers and responsibilities that you supplied. Cut adjectives you did not ask for.
Run a swap test
Give the tool the same experience under a different gender signal, such as a different name, and compare the outputs. If the wording shifts in ways that are not explained by your facts, you have seen the problem firsthand. This is an informal check, not a rigorous audit, but it is quick and revealing.
Target the job, not the stereotype
Paste in the job description and tell the AI to match your real skills to its requirements. Specific instructions leave less room for generic assumptions.
Be careful with AI salary advice
The coverage flags salary guidance as one area where biased recommendations could matter. If you ask an AI what to request in a negotiation and its answer is shaped by stereotypes, a small difference in the number could follow you for years.
Do not rely on one chatbot's figure. Cross-check it against published salary ranges, postings that list pay, and what people in your field tell you. Treat the AI's number as one data point.
What employers should do
If you screen candidates or work in HR, the study's implications are more pointed. Per the reporting, the findings challenge the assumption that automation automatically makes hiring more objective.
Based on the study's core finding, here is what follows:
- Test by occupation, not just overall. An average fairness score can conceal penalties in specific jobs, so audit results role by role, especially in gender-skewed fields.
- Do not rely on gender-neutral prompts alone. The study's framing suggests that simply telling a model to be neutral is not a sufficient safeguard, and that audits need to go further.
- Keep humans accountable. When a tool influences who gets an interview or what a hire is offered, someone should be able to explain and override its output.
Small biases matter because of scale. One skewed resume suggestion is a nuisance. The same suggestion repeated across thousands of applications becomes a barrier.
The bottom line
AI can be a helpful drafting partner, but it is not a neutral one by default. Before you send anything, read the output as a skeptic, test it with a swap, and verify any salary number elsewhere. If you are on the hiring side, ask for results broken down by job type, not one flattering average.