Bias in Hiring: The 11 Most Common Types, How Each Shows Up, and How to Reduce Them

Bias in hiring is not a character flaw of the recruiter. It's a predictable feature of how human brains and algorithmic systems decide under uncertainty. Denying it exists doesn't reduce it, it just guarantees it keeps operating invisibly.

This article is a practical catalog: the 11 most common biases in US hiring decisions today, how each shows up concretely in the funnel, and the practices that actually reduce their impact. No promise to eliminate, a promise of diagnosis and tools.

#Classic human biases

#1. Confirmation bias

A recruiter forms an impression in the first 30 seconds of an interview (or the first read of a resume) and the rest of the conversation serves to confirm that impression. Questions become a search for evidence of the initial hypothesis.

How it shows up: the candidate "passed" the screen if they felt comfortable at the start; "didn't pass" if they produced noise early.

How to reduce:

• Structured interviewing with the same questions for everyone
• Written decision before discussing with colleagues
• Closed rubric instead of "I thought it went well"

#2. Halo (and horns) effect

A single salient positive (or negative) characteristic spreads across the entire evaluation. A candidate from a top university gets rated as better at everything, communication, reasoning, ownership, without evidence.

How it shows up: "She went to MIT, she must be great at distributed systems." Or: "He has an 18-month gap on his resume, must be generally disorganized."

How to reduce:

• Evaluate competencies separately with independent rubrics
• Don't aggregate into a single "overall impression"
• Blind technical tests (no name, no school)

#3. Affinity bias

A tendency to prefer candidates who resemble the recruiter (same social background, same school, same hobby, same personality). The hardest bias to remove because it feels like "chemistry" or "cultural fit."

How it shows up: "I liked her, she'll fit in with us." Often translates to "she resembles me."

How to reduce:

• Interview panels with background diversity
• Separate "cultural fit" (shares explicit company values) from "social fit" (resembles me)
• Document the cultural-fit criteria in writing before the interview

#4. Anchoring bias

The first number or first piece of information shapes everything afterward. A great resume seen first becomes the anchor for the next 20 resumes.

How it shows up: "That's the strongest one in the queue", usually it's the one closest to the first you saw.

How to reduce:

• Evaluate candidates against the rubric, not against each other
• Randomize evaluation order
• Take pauses between batches to reset the anchor

#5. Availability bias

We judge probability based on how easily we recall examples. If the last bad dev was a Vue developer, the next Vue developer looks like risk, without data.

How it shows up: "Framework X devs tend to be weak at architecture", opinion based on 2-3 cases, not on statistics.

How to reduce:

• Decisions based on aggregate data when available
• Document internal heuristics and revisit them periodically
• Replace intuition with objective tests where possible

#6. Status-quo bias

Unconscious preference for candidates who resemble who's already on the team. Actively reduces diversity, even when the company says it values diversity.

How it shows up: Team is all-male, all-Ivy, all backend Java. The next hire who "fits" tends to look the same. "It wasn't intentional", exactly what makes this bias dangerous.

How to reduce:

• Intentional gap analysis ("what's missing from our team?") before opening the role
• Diversified pipeline at the source (not just postings in obvious channels)
• Post-hire demographic metrics, revisited quarterly

#Biases specific to the US market

#7. School-prestige bias

Top US schools (MIT, Stanford, Berkeley, CMU) get disproportionate weight on first read. Candidates from state schools or non-traditional bootcamps compete at a disadvantage even with equivalent performance.

How it shows up: Resume from CMU goes to the top of the queue; equivalent state school goes to the bottom, same role.

How to reduce:

• Technical test before the interview, no school identification
• Decision by measured skill, not by institution name

#8. Coastal / geographic bias

A subtle but documented bias toward candidates in coastal tech hubs. Even for remote roles, candidates listing San Francisco or NYC often get faster responses than candidates from the Midwest or South.

How to reduce:

• Blind initial-read evaluation (no city, no state)
• Geographically diverse pool of candidates
• Approval-rate metrics by region, revisited periodically

#9. Technical-gender bias

Persistent in technical roles. Resumes with male names are read as having technical skill while resumes with female names are read as having "soft" skill. The bias is documented in peer-reviewed studies (Moss-Racusin et al., 2012, and replications), not theory.

How to reduce:

• Blind screening (no name) on first read
• Interview panels with women's participation
• Standardized technical test before behavioral interview

#Algorithmic biases

#10. Inherited bias from historical dataset

An AI model trained on past hiring data learns patterns, including ones nobody wants to replicate. Famous case: Amazon (2018) trained a screening model on 10 years of historical data and discovered it penalized resumes containing the word "women's." Model decommissioned.

How it shows up: AI screening platform "strangely" filters more candidates of demographic X than Y. Wasn't explicitly programmed, it's inherited.

How to reduce:

• Model bias audit with synthetic testing across demographics
• Annual reassessment with balanced dataset
• In regulated jurisdictions (NYC Local Law 144, Illinois AIVIA, Colorado SB205): mandatory annual audit

#11. Resume-optimization bias

Systems that weight keyword match favor candidates who learned to optimize resumes for ATS, not the strongest technically. Creates a meta-bias against those who write concise, focused resumes.

How to reduce:

• Move the filter from "reading resumes" to "measuring skill"
• Platforms like NORT attack this meta-bias via test-based Career Score, not resume parsing

#How to measure if you have bias (and how much)

Stating good intent isn't enough. Four practical tests:

#Test 1: Demographic-rate analysis

Compare advance-rate by stage across gender, race, age, school. If the rate drops more for one group between stage 2 and stage 3, bias is operating at that stage.

#Test 2: JD audit

Does the job description use gender-coded words ("aggressive," "rockstar," "ninja", male-coded; "support," "nurture," "collaborate", female-coded)? Tools like Textio and Gender Decoder help.

#Test 3: Controlled blind evaluation

Apply the same rubric to identified and de-identified resumes (name, photo, school, address). Different outcome is direct proof of bias.

#Test 4: Diversity of the final pool

If the top 10% of candidates by skill has demographic X but the top 10% of hires has another demographic, bias lives in the funnel, not the pool.

#What does NOT reduce bias (despite sounding good)

• "Unconscious bias training" without process change: recent research shows short-term effect with no sustained behavior change
• Declared quota without criteria change: generates resentment without addressing the cause
• "Belonging culture" without data: slogan without instrumentation changes nothing
• AI as automatic solution to human bias: can trade one bias for an algorithmic one

#Regulation in the US

A growing body of US law governs hiring assessment tools:

• NYC Local Law 144: Automated Employment Decision Tools (AEDT) used for NYC-based positions require annual bias audit, candidate notice, and disclosure
• Illinois AIVIA: Artificial Intelligence Video Interview Act requires disclosure and data retention limits
• Colorado SB205 (effective Feb 2026), broader AI accountability framework touching hiring
• California AB-2930 framework: disclosure and bias audit requirements
• EEOC guidance: automated tools fall under Title VII disparate-impact analysis regardless of state location

Companies operating across states should assume an audit is required somewhere in the stack.

#NORT's approach

NORT attacks bias via three architectural mechanisms:

1. Filter by objective criteria over a pre-evaluated pool: score by measured skill, measured language, validated Big Five. The resume becomes peripheral.

2. Portable assessment: the candidate completes the process once; the company doesn't have an opportunity to filter by name, photo, or school on first read.

3. Transparency to the candidate: results are visible to the candidate, so automated decisions carry accountability.

Doesn't eliminate all human bias (the final interview still exists), but moves the funnel's bottleneck to where it's most defensible: skill measurement.

#Frequently asked questions

#Can I eliminate bias entirely?

No. You can significantly reduce it, research shows structured process + blind initial evaluation + clear rubric reduces bias 30-60% across demographics. Total elimination would require removing human decision, which creates other problems.

#Can a company legally set quotas?

In the US, hard quotas in private hiring are generally unlawful under Title VII; programs targeted to underrepresented groups must avoid quota structures. Outreach and pipeline diversification are legal; rigid demographic quotas are not.

#Is bias audit required in the US?

In multiple jurisdictions yes (NYC, Illinois, Colorado, California framework). EEOC enforcement applies federally. Companies operating across states should assume audit is required somewhere in their stack.

#Does AI reduce or amplify bias?

It can do either. A model trained on balanced data, with validated instruments (such as Big Five) and regular audit reduces. A model trained on internal history without audit amplifies.

#Is "cultural fit" disguised bias?

It can be. Cultural fit is defensible when it means "shares explicit documented company values" (e.g., a company that values radical honesty). It's problematic when it means "resembles me socially." The difference is written-criteria clarity.

#TL;DR

• Hiring bias isn't a character flaw, it's a predictable feature of decision-making under uncertainty
• The main ones: confirmation, halo, affinity, anchoring, availability, status quo + US-specific biases (school prestige, coastal/geographic, technical-gender) + algorithmic biases (historical dataset, resume optimization)
• Reduction requires structured process, blind evaluation where possible, clear rubric, demographic metrics
• AI can reduce or amplify, depends on audit
• US regulation (NYC LL 144, Illinois AIVIA, Colorado SB205, EEOC) increasingly mandates bias audit and disclosure
• NORT attacks via objective filter on a pre-evaluated pool, portable assessment, and transparency

---

#Related resources

• AI Resume Screening: how it works and where it fails
• The Big Five in hiring
• How to evaluate a tech candidate without interviewing
• Glossary: ATS
• Glossary: Smart Match

---

Updated May 16, 2026. Comments or corrections: [email protected].