AI in Hiring: Benefits, Bias Risks and How to Get It Right

AI is reshaping how companies find and hire people. It can screen hundreds of applications overnight, schedule interviews without a single email chain and surface candidates a recruiter might never have found manually. For businesses dealing with high-volume hiring or stretched HR teams, that efficiency is genuinely valuable.

But AI in hiring also carries real risks. When the tools you use to make decisions about people are trained on biased data, or designed without adequate oversight, they can screen out qualified candidates based on factors that have nothing to do with the job. And in the UK, the employer is legally responsible for the outcome, not the software vendor.

Here, we explain how AI is used across recruitment, where bias enters the process, what the evidence says and what you can do about it.

What is AI in hiring?

AI in hiring refers to the use of machine learning, natural language processing and data-driven algorithms to automate or support decisions across the recruitment process. It’s not a single tool. It’s a category of technology that shows up at almost every stage, from the moment a job description is written to the moment an offer is made. For a deeper look at how it’s being applied today, see our guide to AI in recruitment.

How AI is used across the hiring process

Here’s where AI typically gets applied:

• Job description writing: AI tools suggest language, flag gendered terms and help write descriptions optimised for search visibility and candidate response rates.
• Candidate sourcing: AI searches job boards, CV databases and professional networks to build candidate lists based on skills, experience and profile signals.
• Resume screening: Algorithms rank and filter applications against defined criteria, often before a human sees a single CV.
• Interview scheduling: Automated tools manage calendars, send invites and eliminate the back-and-forth that slows down coordination.
• Video interviews and assessments: Some tools record video responses and use AI to score candidates on verbal and non-verbal cues, including tone, word choice and in some cases facial expression.
• Final-stage decision support: AI can flag high-potential candidates, predict performance based on historical data, or generate scoring summaries for hiring managers.

Why employers are adopting AI in hiring

The pull towards AI in recruitment comes down to three things: volume, cost and speed. Hiring teams are dealing with more applicants than ever and the manual work involved in screening, scheduling and coordinating interviews doesn’t scale well.

The pressure is real. According to Employment Hero’s own research, 3 in 4 business leaders say recruitment is a challenge and 95% of SMEs report challenges managing employment and HR processes, with recruitment ranking as the single most common pain point. Against that backdrop, AI’s appeal is understandable.

According to the World Economic Forum, 88% of companies now use AI for initial candidate screening. A 2026 Boterview report put overall AI adoption in recruitment at 87%. For businesses under pressure to hire quickly and control costs, these tools offer a clear efficiency case.

That case is real. But it doesn’t tell the full story.

The benefits of using AI in hiring

Used well, AI recruitment tools can genuinely improve the hiring process for both employers and candidates.

Faster screening and shortlisting

Manual CV screening is slow. A recruiter reading 200 applications for a single role can spend hours on a task that AI completes in minutes. For high-volume roles, this time saving is significant. Some estimates put speed-to-hire improvements as high as 75% when AI screening is introduced, though results vary significantly by context and role type. The upside for HR teams is clear: less time on admin means more time for the conversations that actually require human judgment.

There’s also an accuracy argument. Employment Hero research found that 6 in 10 UK businesses have had issues making the wrong hiring decision. Faster screening only delivers value if it also helps identify the right people, not just more people. See how Employment Hero tackles applicant overload with AI recruitment to understand what that looks like in practice.

More consistent candidate evaluation

Human reviewers are inconsistent. The same CV reviewed by different people can receive very different assessments, depending on the reviewer’s mood, workload, familiarity with the role, or unconscious preferences. AI applies the same criteria to every application in the same way. In theory, that consistency can reduce the variability that allows individual bias to creep in at the screening stage.

Improved candidate experience

The scale of candidate frustration with current hiring processes is significant. Employment Hero research found that 8 in 10 UK workers have applied for a job and heard nothing back and 54% say that applying and receiving no response is the single most frustrating part of job searching. More broadly, 6 in 10 say the hiring process actively discourages them from looking for new roles.

AI tools can meaningfully address this. Automated acknowledgement emails, status updates and 24/7 self-service scheduling create a more responsive experience without adding to recruiter workload. When candidates know where they stand, they’re more likely to stay engaged with the process. For more on keeping candidates engaged throughout hiring, see our candidate engagement strategies guide.

To see how AI-powered hiring tools can improve your recruitment process, take a look at Employment Hero’s recruitment solutions.

What is AI bias in hiring?

AI bias in hiring is what happens when an algorithm produces systematically unfair outcomes for certain groups of candidates. It’s not usually the result of someone programming discrimination into a system. It happens because AI learns from data and that data reflects the world as it was, not as it should be.

When a model is trained on historical hiring records, it learns what past successful hires looked like. If those records reflect decades of underrepresentation, the model will reproduce that underrepresentation. It’s not making a moral judgment. It’s doing exactly what it was designed to do: spot patterns and replicate them.

Where does AI bias in hiring come from?

There are several root causes:

• Biased training data. If an algorithm learns from the hiring decisions of the last decade, it inherits all the bias embedded in those decisions. Historical underrepresentation of women in tech, or Black candidates in senior roles, gets encoded as a signal about what good candidates look like. This is closely related to the problem of unconscious bias in recruitment, which shapes hiring outcomes long before AI enters the picture.
• Biased algorithm design. Even when training data is cleaned up, the choice of what variables to include or exclude can introduce bias. Proxies like postcode, name, or university attended can correlate strongly with protected characteristics.
• Feedback loops. When AI systems learn continuously from outcomes, they can reinforce their own bias. If the model already tends to favour certain candidates and those candidates get hired, the model interprets that as confirmation it was right.

Real-world examples of AI bias in hiring algorithms

The risks here aren’t theoretical. They’ve already caused real harm.

Amazon, 2014-2017: Amazon built an internal AI tool to score and rank job applicants. By 2015, the team discovered it was systematically discriminating against women. The model had been trained on CVs submitted to Amazon over the previous decade, most of which came from men. It learned that male candidates were preferable. It penalised CVs that included the word “women’s,” downgraded graduates of all-women’s colleges and favoured masculine-coded language. Amazon scrapped the project in early 2017.

HireVue’s facial analysis controversy: HireVue’s video interview platform used AI to assess candidates’ cognitive ability, emotional intelligence and psychological traits from recorded interviews, analysing facial expressions, speech patterns and word choice. Privacy advocates, civil rights groups and AI researchers challenged the scientific basis for these claims and raised specific concerns about lower accuracy for darker skin tones, as well as disadvantage for non-native speakers, neurodivergent people and candidates with disabilities. In January 2021, HireVue announced it would discontinue facial analysis, acknowledging the technology “wasn’t worth the concern.” In March 2025, the ACLU filed a complaint against Intuit and HireVue after an Indigenous deaf applicant was automatically rejected following an AI interview.

These cases illustrate why AI tools that appear objective can still produce discriminatory outcomes.

Gender and racial bias in AI hiring tools

The academic evidence on this is substantial and growing. A 2025 peer-reviewed study published in PNAS Nexus by researchers at the University of Hong Kong and the Chinese Academy of Sciences tested five large language models, including GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Flash and Llama 3-70B, across approximately 361,000 fictitious CVs. Only names were varied to signal gender and race.

Every single model awarded higher scores to female candidates regardless of race, while most penalised Black male candidates compared to identical white male candidates. The bias was consistent across all five models, from four different AI companies, suggesting the problem is embedded in how these systems are trained broadly rather than being the result of any single vendor’s decisions.

A separate University of Washington study found that AI resume screeners preferred white-associated names 85.1% of the time and Black-associated names only 9% of the time.

The VoxDev summary of the PNAS research flags a specific concern about intersectionality: bias doesn’t operate neatly along single dimensions. Candidates from overlapping minority groups face compounded disadvantage that single-axis regulatory frameworks aren’t designed to catch.

AI bias in the hiring process: Where does it show up?

Bias doesn’t sit in one place. It can enter at every stage of the recruitment funnel.

Bias in resume screening and candidate ranking

This is where AI bias has the most documented impact. When screening algorithms are trained on the characteristics of past successful hires, they systematically favour candidates who resemble those hires and screen out equally qualified people who don’t fit the pattern. The algorithm isn’t asking whether a candidate can do the job. It’s asking whether they look like someone who has done a similar job before and that’s a very different question.

Bias in job descriptions and candidate targeting

The bias can start before a single application arrives. AI tools that write or optimise job descriptions can introduce gendered language, favoured by the models based on which job ads historically attracted the most engagement. Similarly, programmatic job ad targeting can use demographic signals to reach particular audiences, narrowing the candidate pool before hiring has even begun. BSR’s research on AI in hiring found that these tools can perpetuate bias “in surprising and insidious ways” precisely because employers often don’t know where the filtering is happening. Reviewing your ethical recruitment principles is a good starting point for identifying where your own process may be vulnerable.

Bias in video interview and assessment tools

Facial expression analysis and speech pattern scoring tools raise particular concerns. They can disadvantage candidates with accents outside the model’s training data, neurodivergent candidates who communicate differently, or candidates with speech-related disabilities. When a tool scores confidence or warmth based on facial movement, it may simply be measuring how well a candidate’s presentation style matches the model’s idea of what confidence looks like, which is itself a culturally specific construct.

Can AI reduce bias in hiring?

It’s a fair question and the honest answer is: sometimes, but not by default.

There are genuine scenarios where removing subjective human judgment from early-stage screening can produce fairer outcomes. If a hiring manager’s unconscious preferences are causing consistent variability in how CVs are assessed, a well-designed algorithm applying consistent criteria could reduce that variability. Structured, criteria-based evaluation applied uniformly is a proven principle of fair assessment.

But “objective” and “fair” are not the same thing. An algorithm that applies the same criteria to every candidate is objective. Whether those criteria are themselves fair is a separate question entirely. As the BSR points out, an AI tool “marketed as objective is not the same as a tool that is fair.”

The PNAS Nexus research found that the AI models tested appeared to have overcorrected for some biases while introducing others. Attempts to reduce gender bias in one direction created new patterns of disadvantage along racial lines. Bias doesn’t disappear when it’s addressed in isolation; it shifts.

The conclusion that follows is practical: AI can assist in reducing certain forms of human bias at specific points in the process, but it doesn’t eliminate bias overall. Human review remains essential at every key decision point. AI should inform hiring decisions, not make them.

To explore how AI-enhanced HR can work with human oversight built in, see Employment Hero’s AI-enhanced HR solutions.

How to reduce AI bias in your hiring process

Here are practical steps to audit and reduce your bias risk:

1. Audit every tool you use. That includes third-party tools your vendor uses. Ask each provider how their model was trained, what data it was trained on and what bias testing they’ve carried out. If they can’t answer those questions, treat that as a red flag.
2. Conduct a pre-deployment review. Before rolling out any new AI hiring tool, run an ethics and impact assessment. The UK government’s Responsible AI in Recruitment guide (published March 2024) sets out a framework for doing this.
3. Test outcomes across protected groups. Look at the outputs of your AI tools across gender, race, age and disability. If any group is being screened out at a significantly higher rate, investigate before assuming the algorithm is working correctly.
4. Keep humans in key decisions. AI should be a filter and an assistant, not a decision-maker. Ensure a human reviews AI-generated shortlists and that candidates flagged for rejection by AI are genuinely assessed, not just passed through. Our guide on how to hire the right person covers how to build hiring manager accountability into your process.
5. Tell candidates when AI is being used. Transparency isn’t just good ethics, it’s increasingly a regulatory expectation. Candidates should know if they’re being assessed by an algorithm and should have access to a human review process.
6. Revisit and retest regularly. Bias in AI systems can evolve, particularly if models learn continuously from new data. Schedule regular bias audits rather than treating initial testing as a one-off exercise. This is necessary even if the model is not continuously learning, as changes in the job market, applicant demographics, or the job role itself can lead to Model Drift or Concept Drift, causing a once-fair system to become discriminatory over time.
7. Check for Algorithmic Mitigation Techniques. Ask vendors what technical debiasing techniques are being applied. These can include pre-processing (adjusting training data), in-processing (modifying the model during training), or post-processing (adjusting the final output scores). Ensure technical mitigation complements human oversight.

For more on how AI is reshaping the world of work, see our guide to AI in the workplace.

AI in hiring and the law: What employers need to know

Using an AI tool doesn’t transfer your legal responsibility to the tool’s vendor. If your hiring process produces discriminatory outcomes, you carry the liability.

The Equality Act 2010 prohibits discrimination on the basis of nine protected characteristics: age, disability, gender reassignment, marriage and civil partnership, pregnancy and maternity, race, religion or belief, sex and sexual orientation. Critically, intent doesn’t matter. Indirect discrimination, where a practice appears neutral but produces a disproportionate disadvantage for a protected group, is still unlawful. An algorithm that screens out women at a higher rate than men is creating indirect sex discrimination, regardless of whether anyone intended that outcome.

The Information Commissioner’s Office published audit findings from reviews of AI recruitment tool providers and issued 296 recommendations and 42 advisory notes. Auditors found some tools were allowing recruiters to filter candidates based on protected characteristics, that providers were failing to treat inferred demographic data as special category data under UK GDPR and that vast candidate databases were being built through social media scraping without adequate consent. Before using any AI recruitment product, the ICO says employers should confirm the provider has completed a Data Protection Impact Assessment and that candidates are informed AI is being used to evaluate them. If you’re unsure how UK GDPR applies to the personal data you collect during recruitment, our GDPR compliance guide covers the core principles.

The Equality and Human Rights Commission confirmed in April 2024 that AI in recruitment is one of its key focus areas and that AI poses risks of breaching both the Equality Act 2010 and the Human Rights Act 1998.

The UK has no single AI law yet. The current approach uses sector regulators applying a principles-based framework. But legislation is coming. A UK AI Bill was announced in late 2024 and the ICO is preparing updated recruitment-specific guidance. The direction of travel is clear: the regulatory environment for AI in hiring is tightening.

For comparison, in the US, the EEOC settled its first AI hiring discrimination case in 2023, with iTutorGroup paying $365,000 after its software automatically rejected women over 55 and men over 60. In May 2025, a US federal court certified the first AI bias class action against Workday. These cases are a signal of where UK enforcement is likely to move. For a clearer picture of what employment tribunal proceedings can cost a UK employer, see our breakdown of employment tribunal costs.

AI-powered hiring, built responsibly

AI in hiring works best when it’s built responsibly. Employment Hero’s AI is designed to help you hire smarter and faster, cutting through admin and surfacing the right candidates, without replacing the human judgment that fair hiring depends on. If you want to see what that looks like in practice, explore Employment Hero’s AI.