Day 12: Screen CVs Against a Job Description

The concept

CV screening is where many hiring decisions are quietly shaped.

By the time candidates reach interview, the pool has already been filtered. If the screening process is rushed, vague, or overly dependent on familiar career patterns, strong candidates may never get the chance to show what they can do.

AI can support screening by applying a consistent framework. But this only helps if the framework is fair and job-relevant. If the criteria are biased, AI will apply bias consistently.

Plain English

Build the criteria before reviewing candidates. Then score evidence, not familiarity.

Start with the job description

Do not begin by pasting CVs into AI. Begin with the role.

Ask AI to review the job description and identify:

• must-have criteria
• desirable criteria
• vague language
• inflated requirements
• potentially biased signals
• missing evidence requirements

This step matters because the job description often carries assumptions. "Strong academic background" may be unnecessary. "Polished communication" may be vague. "Culture fit" may be risky. "Five years of experience" may be a proxy for complexity that should be described directly.

If the criteria are wrong, the screening will be wrong.

Use a simple rubric

Screening does not need an elaborate scoring system. It needs a consistent one.

A simple scale can work:

• 0: no clear evidence
• 1: partial or indirect evidence
• 2: strong direct evidence

For each criterion, define what evidence counts. This prevents drift.

For example, if the criterion is stakeholder management, strong evidence might include leading cross-functional projects, managing competing priorities, or influencing senior stakeholders without direct authority. It should not simply mean "worked at a famous company."

Anonymise where practical

Anonymisation can reduce some bias, but it is not perfect. CVs may still reveal location, career path, school, or employment gaps. Still, removing names and unnecessary personal details can help the reviewer focus on evidence.

Be careful with data privacy. Follow your organisation's recruitment and AI policies before sharing candidate information with any tool.

If in doubt, anonymise more.

Watch for non-linear careers

Traditional screening often penalises candidates who do not follow familiar paths.

Examples include:

• career breaks
• sector changes
• self-employment
• caregiving gaps
• military transition
• international experience
• portfolio careers
• returners

These patterns should not automatically count against a candidate. The question is whether the CV shows evidence of the required capability.

AI can flag unusual patterns for human review, but HR must make the judgement.

Do not outsource the decision

AI can score against criteria, but it should not make the hiring decision alone.

Human review is needed to:

• approve the criteria
• check fairness
• review borderline candidates
• identify transferable experience
• protect against over-reliance on wording
• consider reasonable adjustments
• ensure legal compliance

Use the model to create structure and consistency. Keep accountability with the hiring team.

Separate minimum criteria from ranking criteria

Screening often becomes unfair when every desirable feature is treated like a requirement.

Minimum criteria answer:

Can this person plausibly do the role with reasonable onboarding?

Ranking criteria answer:

Among qualified candidates, who shows the strongest evidence for this role now?

If you mix these two categories, you may exclude candidates who meet the bar because they lack preferences that are not essential. This is especially common when hiring managers add every nice-to-have into the job description.

Ask AI to separate:

• must-have evidence
• desirable evidence
• evidence to probe in interview
• evidence that should not be used

This keeps the shortlist fairer and easier to explain.

Use evidence notes, not just scores

A score without evidence is not enough. Two candidates may both score 2 on a criterion for different reasons, and the panel needs to understand the difference.

For each score, capture:

• the evidence from the CV
• whether the evidence is direct or inferred
• what is missing
• what should be probed later

This creates an audit trail. It also helps the hiring manager avoid re-screening from scratch.

Be careful with prestige signals

Prestige can quietly distort screening.

Brand-name employers, elite universities, polished formatting, and familiar job titles may influence reviewers even when they are not part of the criteria. Sometimes they are relevant. Often they are not.

If a criterion is "has managed high-volume operations," then the evidence is the work, not the employer brand. If a candidate did comparable work in a smaller organisation, that may be just as relevant.

AI should be instructed not to use prestige signals unless they are directly connected to the role requirement.

Review the rejected pool

Before finalising, review a small sample of rejected candidates.

Ask:

• Did any candidate fail because of an unclear criterion?
• Did career breaks or non-linear paths get penalised too heavily?
• Did the rubric overvalue familiar companies or titles?
• Did any candidate meet must-haves but lose out on nice-to-haves?
• Are there patterns in who was screened out?

This review does not need to be slow. It is a fairness check that can catch problems before interviews begin.

Turn screening into interview preparation

A useful screening process does not end with a shortlist. It should inform the interview.

For each shortlisted candidate, ask:

• Which criteria are strongly evidenced?
• Which are unclear?
• What transferable experience needs probing?
• What risk should the interview test?
• What strength should the panel explore?

This keeps interviews targeted. It also reduces the habit of asking every candidate generic questions that do not address the actual evidence gaps.

Protect candidate data

CVs contain personal information. Before using AI in screening, follow your organisation's approved tools, privacy policies, and candidate notice requirements.

Where possible:

• anonymise names
• remove contact details
• remove unnecessary demographic indicators
• avoid uploading sensitive data to unapproved tools
• keep scoring outputs in the right recruitment system
• document human review

Efficiency should not come at the expense of candidate trust.

Today's practice

Choose one role. Run Step 1 only. Review the criteria before adding CVs.

Ask:

1. Are the must-haves truly essential?
2. Could any criterion unfairly narrow the pool?
3. Does the rubric define evidence clearly?
4. How will we handle non-linear paths?
5. What candidate data should be removed before scoring?

By the end, you should have a screening framework that is clearer, fairer, and easier to defend.