It is a fair question and an important one. If you are considering AI CV screening software, the concern about bias is probably somewhere on your list. Given the widely reported issues with AI bias in other contexts, particularly Amazon's now-scrapped recruitment AI that was found to penalise CVs containing the word "women's," scepticism is reasonable.
The short answer is that AI CV screening can be biased, but it does not have to be, and the risk is often lower than the bias that already exists in manual screening. Here is what you actually need to know.
Where AI Bias Comes From
AI systems learn from data. If the data they are trained on reflects historical patterns of discrimination, the AI can learn to replicate those patterns. This is what happened with Amazon's tool: it was trained on years of historical hiring data from a male-dominated industry, and it effectively learned to prefer male candidates.
This is a real risk in AI systems that have been trained on historical hiring outcomes, particularly those built by large tech companies processing millions of historical applications.
The risk is considerably lower in AI tools that are not trying to predict "good hires" based on historical data, but are instead matching candidate experience and skills against specific, defined job criteria. If the tool is comparing what a candidate has done against what a role requires, and those criteria have been set by a human recruiter with a clear brief, there is no historical hiring data involved in the scoring. The AI is doing pattern matching against the job description, not predicting success based on what past hires looked like.
The Bias That Already Exists in Manual Screening
Before concluding that manual screening is safer, it is worth being honest about what the research says about human bias in CV screening.
Studies have consistently found that identical CVs with different names at the top receive different levels of interest from human screeners. CVs with traditionally white British names receive significantly more positive responses than identical CVs with names that read as South Asian or Black. Gender bias, age bias, and educational institution bias have all been documented in manual recruitment screening.
Human screeners are also subject to fatigue, which as noted elsewhere tends to make bias worse rather than better. A CV that lands at number 250 in a pile of 300 is being assessed by a different version of the recruiter than the one who read number 10.
This is not to excuse AI bias. It is to say that the relevant comparison is not AI versus a perfectly objective human, but AI versus a tired human with unconscious preferences.
What Good AI Screening Actually Does
The best AI screening tools focus on matching candidates to defined role criteria: specific experience, relevant skills, qualifications, and career history. When the scoring is based on that kind of explicit, role-specific matching, the name, age, gender, and appearance of the candidate are not factors in the result.
That is actually a significant advantage over manual screening in terms of consistency. Every candidate is assessed against the same criteria with the same level of attention, at 2am on a Sunday or 9am on a Monday.
The tool is only as good as the criteria set for it, which means the recruiter's judgement still matters. If the job criteria you input reflect biased assumptions about what a good candidate looks like, the output will reflect that. The responsibility for setting fair, relevant criteria sits with the person using the tool.
Practical Steps for Using AI Screening Fairly
Focus criteria on demonstrable experience and skills. The more clearly the criteria relate to what the role actually requires, the less room there is for irrelevant factors to influence the result.
Review the output with professional judgement. AI screening should narrow the field, not make the final decision. Your role as the recruiter is to review the shortlist, apply context that the AI cannot access, and make informed decisions about who to put forward.
Check for unexpected patterns. If you run a batch through and notice that certain types of candidates are consistently ranking low in ways that do not match your professional expectation, that is worth investigating.
Keep humans in the decision loop. No AI screening tool should be the only thing standing between a candidate and rejection. The final shortlist should always involve a person who understands the role, the client, and what good looks like.
The Bottom Line
AI CV screening is not inherently more biased than manual screening, and in many cases it is significantly less so. The key is using tools that score candidates against defined role criteria rather than predicting success based on historical data, and applying your own professional judgement to the output.
Used responsibly, AI screening can actually make your process fairer by applying consistent criteria to every candidate, removing the fatigue effect that degrades human judgement at high volume, and surfacing strong candidates who might otherwise have been missed.
Lucuma's screening is built around role-specific matching, comparing what each candidate has done against what the role requires. There is a 14-day free trial with no card required if you want to see how it works in practice.