Your Career Gap Is a Red Flag to AI — Even When It Shouldn't Be

What the AI Is Actually Screening For

Seventy-nine percent of hiring managers now use AI somewhere in their recruitment process. Most candidates know this. What they don't know is what these tools are actually screening for — and it's not just skills and experience.

New research shows that AI hiring tools routinely use proxy variables — career gaps, hobbies, language patterns, even the specific words on your CV — to infer things like gender, ethnicity, and age. Then they penalise you for them. Not deliberately. Not visibly. But consistently.

What Are Proxy Variables?

A proxy variable is an indirect indicator that correlates with a protected characteristic. In recruitment AI, this means the system doesn't need to know your gender to discriminate by gender. It just needs data that statistically correlates with it.

A comprehensive study published in ScienceDirect found that career gaps are one of the most significant proxy variables for gender in AI recruitment. Women are disproportionately the primary carers in society, which means gaps in employment history — for maternity, childcare, or eldercare — act as a statistical stand-in for being female.

The AI doesn't “think” women are less qualified. It learns from historical hiring data that candidates with gaps got hired less often, and optimises for that pattern. The effect is the same: systematic disadvantage.

It Goes Deeper Than Gaps

Career gaps are the most obvious proxy, but they're far from the only one. Research from VoxDev found that AI hiring tools use a range of seemingly neutral signals to infer protected characteristics:

• Hobbies and interests: Activities listed on CVs can correlate with gender (e.g., certain sports, volunteering types)
• Language patterns: The specific words and phrasing you use carry statistical gender signals. Women and men tend to describe achievements differently — collaborative vs. individual framing, for instance
• Word embeddings: AI systems represent words as mathematical vectors. Research has found that the vector for “computer programmer” sits closer to “man” than “woman” in the embedding space — meaning the AI treats programming experience as slightly more male-associated, even though the term is gender-neutral

The same study found that large language models favoured white-associated names 85% of the time when ranking CVs, and never once favoured Black male-associated names over white male-associated names. The bias isn't subtle. It's structural.

Why This Matters Now

If AI recruitment tools were niche, this would be an academic concern. They're not niche anymore.

The Resume Genius 2026 Hiring Insights Report, surveying 1,000 hiring managers, found that 79% of companies now use AI in hiring. As their researchers put it: “The biggest mistake job seekers can make right now is assuming a human reads their resume first.”

Meanwhile, Korn Ferry's 2026 TA Trends report found that 52% of talent leaders plan to add autonomous AI agents to their recruitment teams this year, and AI usage across HR tasks has climbed to 43%, up from 26% in 2024.

The tools are spreading. The bias is baked in. And candidates are starting to notice — 66% of adults now hesitate to apply for roles they know use AI screening.

What Can You Actually Do?

To be honest, there's no magic fix when the system itself is the problem. But there are practical steps that reduce your exposure to proxy-based filtering:

1. Address gaps directly, don't leave them blank

A gap on your CV is worse than a gap with context. If you took time for caregiving, freelance work, study, or anything else — say so. AI systems that flag unexplained gaps are less likely to penalise explained ones, because the pattern they've learned is “unexplained gap = risk.”

For more on how to handle this, our guide to explaining career gaps covers the full approach.

2. Mirror the job description's language

This isn't about keyword stuffing. It's about reducing the linguistic distance between your CV and the role requirements. If the job says “led cross-functional teams,” use that phrasing rather than “collaborated with different departments.” You're not gaming the system — you're reducing the noise that proxy-based scoring can latch onto.

3. Quantify impact wherever possible

Numbers are harder for AI to interpret through a gendered lens than adjectives are. “Increased team output by 30%” carries less proxy signal than “passionate about driving team success.” Concrete metrics give the AI less room to infer things it shouldn't.

4. Be strategic about optional fields

Hobbies, interests, and personal statements are proxy-variable goldmines. If you include them, be intentional. If they don't add genuine value to your application, consider leaving them off. Every data point you provide is another input the algorithm can use — not always in your favour.

5. Don't assume fairness

This is the uncomfortable one. AI recruitment tools are not audited consistently. Outside of a New York City law requiring bias audits, there is virtually no regulatory requirement for companies to check whether their AI screening discriminates. You cannot assume the system is fair. Apply accordingly — cast a wider net, don't put all your weight on a single automated pipeline, and value companies that are transparent about their hiring process.

The Bigger Picture

The promise of AI in recruitment was objectivity. Remove human bias, let the data decide. The reality is that AI doesn't remove bias — it launders it through mathematics. Historical data carries historical prejudice, and proxy variables give algorithms a backdoor to discriminate while looking neutral.

This isn't an argument against AI in hiring. It's an argument for transparency, auditing, and honesty about what these tools actually do. Organisations that report a 48% increase in diversity hiring with aligned AI tools prove the technology can work. But it only works when someone is actively checking for the proxies, correcting the training data, and holding the system accountable.

Until that's the norm rather than the exception, job seekers need to understand that the AI reading their CV isn't neutral. It's making inferences about who you are from signals you probably didn't know you were sending.

Regulators are starting to catch up with this. In the UK, the ICO's audit of AI hiring tools found platforms inferring gender and ethnicity from candidate names, and its draft guidance is now heading toward becoming the legal benchmark for how these tools can use your data.