Screen resumes fairly and without bias

Recruiter · Time: 30 min · HR

Tools that fit this task

• ChatGPT $20/mo
• Claude $20/mo
• HireVue price on request

Prompt

Role: you are an HR partner who assesses candidates for a vacancy on competencies only.
Context: vacancy {{title}}, description: {{key duties}}. Criteria with weights: {{criterion 1, weight}}, {{criterion 2, weight}}, {{criterion 3, weight}}. Signs of a strong match: {{description}}. Below are anonymized resumes of candidates 1..N.
Task: score each candidate from 0 to 100 on each criterion and give a weighted total. Support every score with a fact from the resume. Ignore and leave unmentioned age, gender, marital status, nationality, the school as such and career gaps. For each candidate give three interview questions that test the weak spots.
Format: a table candidate | criterion 1 | criterion 2 | criterion 3 | total | facts from the resume | questions. After the table, separately: what the criteria themselves lack to make the assessment fairer. I make the shortlist decision myself.
Resumes:
{{resumes}}

Steps

1. Write evaluation criteria with weights: must-have skills, relevant experience, extra advantages, and describe what a strong and a weak match looks like for each
2. Anonymize the resumes: remove name, contacts, photo, date of birth, gender and address, keep experience, skills and education
3. Number the candidates and give the assistant the job description, the criteria and the resumes with a prompt for scoring
4. Get a table with scores for each criterion, the reasoning and interview questions
5. Reread the reasoning and discard any conclusion based on indirect signals (graduation years, school names, career gaps)
6. Decide on the shortlist yourself and record why you chose these people

How to check the result

Every score in the table is explained by a fact from the resume, the reasoning mentions no age, gender or other protected characteristics, and a person made the shortlist decision

Pitfalls

• The model marks down career gaps and unfamiliar company names, which is discrimination by indirect signals
• Resumes with names and contacts are personal data: only the anonymized version goes into a public service, and a person makes the hiring decision

Data that stays out of public AI tools

• Candidate CVs with contacts, date of birth and photo: into a public service only the experience without contacts
• Conclusions about a candidate's health, marital status and age: never recorded or analyzed
• Salary expectations and offers of specific people: only in the company system

Other tasks for recruiters

• Build a Boolean search string for candidates
• Write a personal outreach message to a candidate
• Write a job ad and a job description
• Prepare a structured interview
• Score candidates after interviews on the same criteria
• Prepare a candidate profile for the hiring client
• Write a rejection email to a candidate
• Find research studies on a topic
• Write an email to a foreign partner in their language

AI tools for recruiters · Prompt: Screen resumes fairly and without bias