A clear-eyed look at where AI screening goes wrong, why structured AI screening outperforms unstructured human review, and how Criba is built to reduce bias rather than amplify it.
AI candidate screening can encode bias if the underlying model is trained on biased historical data or evaluated against irrelevant criteria. However, structured AI screening — where every candidate answers the same questions and is scored against explicit role criteria — is generally fairer than ad-hoc human review, which is vulnerable to halo effects, name bias, and inconsistency. The outcome depends on how the tool is designed, audited, and used.
Bias in candidate screening is not a new problem — it existed long before AI entered the picture. Resumes are filtered by school prestige, names that sound unfamiliar, or gut feel accumulated over years of pattern-matching. AI tools can inherit these same biases if they are trained on past hiring decisions or allowed to evaluate proxies for protected characteristics. At the same time, the structural nature of AI screening — consistent questions, explicit criteria, documented evidence — creates an auditable process that is far easier to inspect and correct than a recruiter's memory of a phone screen. The key is choosing tools designed with fairness in mind and maintaining human oversight at every decision point.
Criba gives all candidates the same voice interview questions in the same order. There is no chance a recruiter asks a follow-up that reveals a candidate's age, family status, or accent. Consistency is the foundation of fairness.
Scores in Criba are tied to criteria you define for the specific role — communication, relevant experience, problem-solving. A candidate is not compared to a vague archetype of a good hire; they are measured against the job.
Every Pass, Borderline, or Reject label links directly to quotes from what the candidate actually said. Recruiters can read the evidence, challenge the score, and override it. Nothing is hidden inside a model.
Criba produces a ranked shortlist — it does not make a hire or no-hire decision. A human recruiter reviews each result and decides who to advance. AI narrows the pool; people make the call.
Screening in Criba is based on what candidates say in their own words, not on the font of their resume, the school name, or whether a surname looks familiar. The evaluation begins with substance.
Because scores are transparent and tied to stated criteria, teams can review patterns across a hiring cycle, spot anomalies, and adjust criteria or questions before the next role. Bias can be caught and fixed.
Define criteria before you screen
Write down what good looks like for the specific role before any candidate answers a question. Criteria set in advance prevent the goal from shifting based on who you happen to like early in the process.
Use consistent questions for all candidates
Every candidate should face the same questions in the same structure. Tools like Criba enforce this automatically — removing the variability that makes ad-hoc phone screens so prone to interviewer-driven bias.
Review evidence, not summaries
When a score appears, read the underlying quotes before accepting it. Criba surfaces the candidate's exact words alongside each score so you are evaluating the person, not a model's interpretation of them.
Keep humans accountable for the final decision
AI screening should narrow the field and surface evidence — not decide who gets hired. Ensure a qualified human reviews shortlist results and is empowered to override any score that does not hold up under scrutiny.
It can, if the tool uses criteria that correlate with protected characteristics or was trained on historically biased hiring data. Criba avoids this by scoring candidates against role-specific criteria you define, not against a historical dataset of past hires. Evidence quotes let you verify every score for yourself.
Legality depends on jurisdiction and implementation. In general, screening tools that evaluate job-relevant criteria consistently across all candidates are on firmer legal ground than tools that infer personality or aptitude from indirect signals. Criba focuses on role-relevant communication and content criteria. We recommend consulting legal counsel for your specific context.
Yes, in meaningful ways. Structured, criteria-based AI screening removes common sources of human bias such as name recognition, resume formatting preferences, and interviewer mood. Criba gives every candidate an identical experience and scores them against the same bar — fairer than most ad-hoc first-round calls.
Criba's scores are not final decisions. Every result includes direct quotes from the candidate's answers so a recruiter can read the evidence and override the score. Human review is a required part of the process, not an afterthought.
Resume screening infers fit from proxies — school, job title, formatting — which can strongly correlate with demographic factors. Criba evaluates what candidates actually say in a structured voice interview, scored against criteria you set for the role. The input is richer and the criteria more transparent.
Start by reviewing the criteria you set for the role and the evidence quotes attached to borderline or rejected candidates. If you notice a pattern — for example, a criterion that systematically disadvantages a group — you can revise the criteria before the next hiring cycle. The audit trail is always there.