You can reduce bias in resume screening, but you cannot remove it with one trick. The most reliable approach combines written job-related criteria, evidence recorded for each decision, more than one reviewer, limits on information that is not needed for the first pass, and checks on your own results. This guide gives 15 practical steps, with the limits of each.
Key takeaways
- Bias in screening comes from habits of the mind and from the process. Fix both.
- Write job-related criteria first, record evidence and use the same sheet for everyone.
- Hide details that are not needed for the first pass, but know the limits: public pilots show mixed results.
- Use two reviewers on a sample, and check your outcomes for patterns.
- Never use protected characteristics, or close proxies for them, to decide.
Where bias enters screening
| Type | What it looks like in screening | Control |
|---|---|---|
| Similarity bias | Preferring applicants who share your school, hobby or background | Evidence sheet and written criteria |
| Halo effect | One strong item, such as a famous employer, lifts every judgment | Mark one criterion at a time |
| Stereotyping | Assumptions linked to a name, age, gender or nationality | Hide details in the first pass; test criteria for proxies |
| Anchoring | The first CV you read sets the standard for the rest | Calibrate on sample CVs; review in random order |
| Fatigue | Later CVs get less attention | Time boxes and short sessions |
| Confirmation | You look for facts that support a first impression | Record evidence for each criterion before you decide |
These are normal human tendencies, not signs of bad intent. A good process assumes they exist and builds in checks. Discrimination law looks at what you did and the effects, not only at what you meant. The US Equal Employment Opportunity Commission lists protected traits and practices to avoid on its page about prohibited employment policies and practices, and the Acas page on recruitment explains the Equality Act position for Great Britain.
Fifteen practical steps
Steps 1 to 5 happen before you screen, steps 6 to 10 while you screen, and steps 11 to 15 after you screen.
- Write the criteria first. Take three to five job-related must-haves from the role brief. Do not add new ones after you see the pile.
- Test each criterion for proxies. Ask: could this exclude people because of age, sex, race, religion, disability, family status or nationality, without being needed for the work? Rewrite it as a task if so.
- Allow several ways to show a skill. “Customer support work, paid or unpaid” is wider and fairer than “two years in a call centre”.
- Define the evidence. For each criterion, write what counts and what does not, with one example.
- Calibrate reviewers. Have everyone mark the same 10 CVs alone, compare, and fix the wording where they differ.
- Hide what you do not need in the first pass. Names, photographs, addresses and dates of birth are not criteria. Where your system allows, hide them for the first review. See the limits below.
- Use a sheet, not a feeling. Mark yes, no or unclear for each criterion and write the CV passage for each yes.
- Review one criterion at a time. Mark criterion 1 for a batch of CVs, then criterion 2. This reduces the halo effect.
- Use random order and time boxes. Mix the batch so that no group is always read at the end of the day. Take breaks.
- Ask about unclear marks. Send the same neutral question to every applicant with an unclear mark before you decline.
- Add a second reviewer on a sample. Have a colleague re-read 10 percent of the declined applications and a few of the accepted ones.
- Look at near misses. Review applicants who failed only one criterion. If many could do the job, the criterion may be too strict.
- Check patterns, lawfully. Where the law allows you to collect monitoring data, compare pass rates by stage and source. If a pattern appears that you cannot explain by job-related reasons, take advice.
- Give every applicant an outcome. Silence is not neutral. People from groups who face barriers are more likely to give up when they hear nothing.
- Keep records, and delete on schedule. Records show that you used job-related criteria. Data protection law also requires you to keep applicant data only as long as needed.
A simple screening sheet you can copy
One sheet per applicant keeps every decision on the same footing. Use one row per criterion. The evidence column is the most important one: if you cannot paste a passage from the CV, the mark is “unclear”, not “yes”.
| Criterion | What counts | Mark (yes / no / unclear) | CV passage or note |
|---|---|---|---|
| 1. Handles customer queries in writing | Any paid or unpaid role with written replies to customers, members or users | ||
| 2. Uses a ticket or case system | Any named system, or a clear description of logging and tracking cases | ||
| 3. Works to deadlines | A stated target, rota or service level they met | ||
| 4. Can work the shifts listed | Stated availability, or no conflict visible in the CV |
Notice what the sheet leaves out: school name, employer prestige, age, and anything about family. If a criterion does not appear on the sheet, it should not influence the decision. If a reviewer wants to add a reason during the review, pause and agree whether the new reason is job-related. If it is, add it for everyone and re-check the earlier applicants.
A worked example (fictional)
This example is invented to show the method. It is not real data. A small company is hiring a customer support agent and gets 120 applications. The hiring manager writes four criteria and defines the evidence. Two reviewers each mark the same 10 CVs alone. They disagree on three of them, all on criterion 3, “works to deadlines”. The wording was too vague, so they rewrite it as “a stated target, rota or service level they met”, and mark the 10 again. Now they agree on nine.
The reviewers then split the remaining CVs and mark criterion 1 for the whole batch, then criterion 2, and so on. Fourteen applicants fail exactly one criterion. A reviewer looks at those 14 and finds that nine of them fail criterion 2, “uses a ticket or case system”, only because they used a different system with a different name. The team changes the “what counts” column to accept any system, and moves the nine forward. The process found a flaw in the criterion, not in the applicants.
What to avoid
| Common mistake | Why it causes harm | Better practice |
|---|---|---|
| Rejecting for a gap in work history | Gaps can relate to health, caring duties or study, and say little about skill | Ask about skills and recent evidence, not about the gap |
| Preferring “culture fit” | It often means “like us” | Define the behaviours the job needs and look for evidence |
| Using a prestige school or employer as a shortcut | It favours people with access, not people with skill | Mark the skill evidence itself |
| Changing criteria after seeing the pile | It lets first impressions steer the rules | Freeze criteria; record any change for all |
| Scanning quickly and trusting the gut | Fast reading amplifies stereotypes | Use the sheet and a time box |
The limits of hiding information
Many employers hide names and other details in the first pass. This can help, but evidence from public programmes is mixed, so treat it as one tool, not a cure.
The Public Service Commission of Canada ran a name-blind pilot with 27 recruitment processes between April and October 2017. Its final report found that anonymisation reduced the overall rate at which applicants were screened in, from 48.3 percent to 43.3 percent. For visible-minority applicants, the rates did not change significantly (47.3 percent and 46.0 percent). The report notes limits, such as that reviewers knew they were part of a pilot and that removing details also removed information related to skills. See the Public Service Commission of Canada report. A trial in the Australian Public Service, with more than 2,100 public servants shortlisting for a hypothetical role, also advises caution before assuming that blind review improves diversity outcomes. See the Australian government report summary.
The lesson is practical. Hiding information can remove one route for bias, but it can also remove context, and it does not fix unclear criteria. Combine it with the other steps, and measure what happens in your own process. We discuss this in more detail in our guide to blind CV screening.
Do not use protected traits, or close proxies
In screening, do not use a protected characteristic, directly or by a proxy, as a reason to include or exclude. Examples of risky proxies include graduation year (age), postcode (race or class), name (ethnicity or gender), gaps in work history (sex, disability or caring duties) and club memberships (religion or sexual orientation). If you think a requirement is truly needed and has an uneven effect on a group, record why it is necessary for the job and look for a less exclusive alternative. Get advice in such cases.
What differs in the US, UK and EU
| Region | What to know | Where to read |
|---|---|---|
| United States | Federal law prohibits job discrimination on protected grounds, and state and city laws can add more. Rules on automated tools and on monitoring differ by state or city, so check where you hire. | EEOC: prohibited practices |
| United Kingdom | The Equality Act 2010 protects nine characteristics. Data protection rules apply to applicant data, including data you collect for monitoring. | Acas: recruitment |
| European Union | Equal treatment is set by EU directives and national laws, and GDPR governs applicant data. National rules differ, so check each country where you hire. | Check the official national equality body and data protection authority |
Using software without losing control
Some tools hide details, extract CV passages or score applicants. If you use one, remember three rules. A person must make and understand the decision. You must be able to explain what the tool did with each application. And you must check the tool’s effect, because a tool can copy bias from the data it learned from. In the UK and the EU, rules on automated decisions and on the use of AI in hiring may apply. The UK Information Commissioner’s Office explains data protection duties in its guidance on recruitment and selection.
FastHire is being built for this kind of person-led screening. It shows each job requirement next to the exact CV passage that supports it, so a reviewer can check the evidence in seconds. FastHire is coming soon. It does not rank, score or reject candidates. People make every decision.
A one-page checklist
| Check | Done? |
|---|---|
| Criteria written before screening and tested for proxies | |
| Evidence defined, with one example per criterion | |
| Reviewers calibrated on 10 CVs | |
| First-pass details hidden where possible | |
| Sheet used: yes, no, unclear and a CV passage | |
| One criterion reviewed at a time | |
| Unclear applicants asked one neutral question | |
| Second reviewer checked a sample | |
| Near misses reviewed | |
| Every applicant told the outcome | |
| Records kept and deletion date set |
Frequently asked questions
Can unconscious bias training fix biased screening?
Training can raise awareness, but it rarely changes results alone. Changing the process, with written criteria, evidence and checks, has more direct effects.
Should I always hide names on CVs?
It can help in the first pass, but results from public pilots are mixed and it does not fix unclear criteria. Use it with other steps and check your own results.
Is it legal to collect diversity data from applicants?
Often yes, if you follow your country’s rules, keep the data separate from decision records and explain its purpose. Rules differ by country, so check the official guidance.
How many reviewers do I need?
At least two on a sample of decisions. For high-volume roles, calibrate all reviewers first.
What if my results show a pattern I cannot explain?
Do not ignore it. Review the criteria and the process, and get advice from an HR or employment law specialist.
Related guides
- Hiring guides for fair and compliant hiring
- Structured screening: how to compare every applicant on the same evidence
- Screening at volume: a process for 500 applications
- Must-haves vs nice-to-haves: cutting your applicant pile without unfair filters
- All hiring guides for employers
- What you can and cannot ask in a job interview: US, UK and EU
- Blind CV screening: does it work and how to do it
- Adverse impact and the four-fifths rule explained
- Disparate treatment vs disparate impact in hiring
- UK Equality Act: direct and indirect discrimination in recruitment
- How to avoid bias in interviews: 12 practical controls
- GDPR for job applicants: what employers must do
- Criminal record checks in hiring: fair practice by country
- How to run an inclusive sourcing process
- How to run an inclusive job ad check
- Fair hiring for older workers and career returners
This is general information, not legal advice.