100×worker · task analysis

Can AI Screen CVs?

By Hendrik De Winne Last updated: Lees dit in het Nederlands
Partly Partly. AI can scan CVs fast for keywords, education, years of experience, and language skills, then produce a first ranking. It lacks context, can inherit bias from training data, and sometimes discriminates unintentionally on name, age, or career gaps. Use it as a pre-filter. A human reviews every shortlist and every rejection.

How to do it

Start with a clear job description and a short list of hard requirements: degree, years of experience, required languages, specific technical skills. Load the CVs (PDF or Word) together with the job description into an AI tool built for document analysis. Ask it to score each candidate against your criteria and to explain each score in one or two sentences, not just a number.

Use this to sort large applicant pools before anyone reads a single CV by hand. It works best for high-volume roles with clear, checkable requirements (a required certification, a minimum number of years, a specific language pair). For roles where fit depends on judgment, potential, or unconventional experience, treat the AI ranking as a starting point rather than a verdict, and read a meaningful sample yourself, not only the top of the list.

Step by step

  1. Write down the job description and the hard requirements in one clear document.
  2. Load the CVs (PDF or Word) together with the job description into your AI tool.
  3. Give the AI concrete criteria: degree, years of experience, languages, technical skills.
  4. Ask for a table or ranking with a score and a short justification for each candidate.
  5. Read the top of the list yourself and spot-check several lower-ranked CVs.
  6. Decide who gets invited yourself, and document why you deviate from the AI ranking when you do.

Where it breaks down

AI reads the CV literally. A skill listed without evidence or context sometimes gets more weight than it deserves, while relevant experience described in unusual wording gets missed. A candidate with a nonlinear career path, a lateral move into the field, or a gap for caregiving or illness can end up ranked lower than they should, simply because the pattern doesn't match what the model was trained to recognize as a strong CV.

There is also a legal and ethical risk. Models can discriminate unintentionally on name, gender, age, or location if those patterns existed in the training data, and rules on AI use in hiring differ by country and are changing quickly (the EU AI Act, for example, treats some recruitment AI as high-risk). Check the current rules for your sector and jurisdiction before you rely on AI screening at scale. Always have a human check the shortlist and every rejection, and keep a record of the criteria the AI used to select or reject each candidate.

A prompt to start with

Rank CVs against a job description

You are helping me screen job applications. Job title: [job title]. Job description and requirements: [paste job description]. Below are [number] candidate CVs. For each candidate, give: 1) a score from 1-10 against the stated requirements, 2) a one-sentence justification, 3) any requirement they clearly do not meet. Do not infer anything not stated in the CV. Flag any candidate where you are uncertain rather than guessing. CVs: [paste CVs or candidate summaries]

Jobs where this task is shifting

Which of your tasks can AI take over?

The free scan builds your personal task map in 30 seconds.

Run the free task scan

Frequently asked questions

Is it legal to use AI to screen job applications?

In most places, yes, but with conditions that vary by country and are tightening. Some jurisdictions classify automated hiring tools as high-risk and require disclosure, human oversight, or bias audits. You remain responsible for discriminatory outcomes even if an AI tool produced the ranking. Check current employment and AI regulations for your country and industry before using AI screening at scale, and keep records of the criteria used.

Can AI screening remove bias from hiring instead of adding it?

It can reduce some human inconsistency, such as fatigue or mood affecting judgment late in a long stack of CVs, but it does not remove bias by default. Language models can learn patterns from training data that correlate with name, gender, age, or location, and reproduce them at scale. Reducing bias requires deliberate design: removing identifying details before scoring, auditing outputs for disparate impact, and having humans review edge cases, not just trusting the tool.

Which candidates does AI screening miss most often?

Candidates with nonlinear career paths, career changes, employment gaps for caregiving or illness, or experience described in industry-specific wording that doesn't match the job posting's exact terms. AI models tend to reward CVs that mirror the language of the job description closely. Strong candidates who describe their experience differently, or whose value shows up in results rather than keywords, can rank lower than they should.

Sources

This article was drafted with AI assistance and editorially reviewed.