Key takeaways
• Automate the evidence work, keep people on the decisions. AI should find, check, organize and score evidence. People should judge exceptions and make every decision to reject, advance at key stages or hire.
• Most candidates accept AI in parts of hiring but not in the final call. In Pew Research surveys, 71% of US adults opposed AI making final hiring decisions, while only 7% supported it.
• Every screening task fits one of three buckets: automate, assist (AI drafts, a person approves) or keep human.
• Exceptions need rules set in advance: low confidence, anything outside the rubric, accommodation requests and requests for a human all go to a person.
• US rules on AI in hiring are growing. New York City, Illinois and California already regulate it, and existing anti-discrimination laws apply to every employer.
What is AI candidate screening?
AI candidate screening is the use of software, including machine learning and large language models, to find, review, question, assess or rank job candidates before a hiring decision. It can cover resume review, practical screening questions, outreach, scheduling, skills assessments and structured interviews.
The one rule to remember
Use AI to gather, verify and organize evidence about candidates, and keep people responsible for judgment, exceptions and every decision that rejects or hires someone. Most screening problems come from breaking this rule in one of two directions: letting AI quietly filter people out, or keeping people stuck on scheduling and follow-up emails that software handles better.
What AI screening is not
AI screening is not the same as an automated hiring decision. A tool that ranks candidates, drafts scores or flags risks is assisting a decision. A tool that rejects candidates with no person reviewing the outcome is making one. The second is where most legal, ethical and quality problems start.
Why the line between AI and people matters now
Four forces make this question urgent for any team hiring in the US in 2026: volume, candidate trust, legal liability and decision quality.
Volume has outgrown manual screening
Applications per hire have tripled since 2021, to more than 300 per open role, according to Ashby data reported by HR Dive. No recruiting team can read every application closely at that volume. Some automation is no longer optional. The real choice is what to automate.
Candidates accept AI in parts of hiring, not in the final call
Pew Research Center surveyed more than 11,000 US adults on AI in hiring. 66% said they would not want to apply for a job with an employer that uses AI to help make hiring decisions, and 71% opposed AI making the final hiring decision (Pew Research Center). Yet 47% thought AI would do better than humans at evaluating all applicants the same way.
What this tells employers
Candidates value the consistency AI can bring, but they want a person accountable for the outcome. A process that uses AI for structured, consistent evaluation and keeps a named person on the decision matches what candidates say they want. Gartner's 2025 research points the same way: only 26% of applicants trust AI to evaluate them fairly (Gartner).
Legal liability now reaches AI tools and the employers who use them
In Mobley v. Workday, an applicant alleges that AI-driven screening tools discriminated against applicants based on age, race and disability. A federal court let the case proceed as a collective action, and in 2026 it continued to reject motions to dismiss most claims, keeping both the vendor and its customers in focus (Gunderson Dettmer, Spring 2026). Whatever the final outcome, the message for employers is clear: "the software decided" is not a defense. Existing laws on discrimination apply to automated screening, and several states have added their own rules, covered later in this guide.
Structure improves decisions, and AI makes structure easier
Hiring research consistently finds that structured evaluation beats informal judgment. Structured interviews have an average validity of .42 for predicting job performance, compared with .19 for unstructured interviews (SIOP, on Sackett et al., 2022). AI is good at running the same questions and rubric for every candidate. Used well, it makes screening fairer and more predictive. Used badly, it automates bias at scale.
The automate, assist, keep-human framework
Every screening task fits one of three buckets. Automate means software does the task and people spot-check. Assist means AI drafts a recommendation with its reasons and a person approves it. Keep human means a person does the task, with AI providing information at most.
Tasks to automate
• Searching for candidates across platforms: high-volume research work where AI covers far more ground.
• Answering candidate questions about the role and process: factual, repetitive and time-sensitive.
• Confirming practical details (location, timing, pay range, interest): clear answers, with unclear ones going to a person.
• Outreach, follow-ups and scheduling: repetitive coordination that slows teams down.
Tasks where AI assists
• Reviewing applications against the scorecard: AI can rank and explain, and a person reviews before anyone is screened out.
• Structured first-round interviews: AI runs the same questions for everyone, and people set the questions and rubric.
• Scoring skills and interview answers: AI drafts scores with evidence, and a person checks the reasoning.
• Building the shortlist: AI proposes, and the recruiter and hiring manager agree it.
Tasks to keep human
• Rejecting a candidate: a decision about a person needs a named owner.
• Weighing unusual backgrounds and career gaps: non-standard paths do not fit trained patterns.
• Accommodation requests and candidate complaints: these need judgment, flexibility and legal care.
• Final interviews, references and the hiring decision: accountability and relationships stay with people.
Stage-by-stage guide to AI candidate screening
The sections below walk through a typical hiring process and show exactly where AI fits at each stage.
Stage 1: Sourcing
Sourcing is the strongest case for automation. AI can search profiles, code repositories, portfolios and publications far faster than a person, and it can search for evidence of work rather than keywords.
What to automate
• Searching across platforms using the role scorecard
• Pulling evidence of relevant work for each candidate
• Removing duplicates and candidates already in your pipeline
What people should check
Review the first batch of results for every new role. If the list looks narrow, for example all from the same few companies or schools, adjust the search before it shapes the whole pipeline.
Stage 2: Application review
Application review is where most AI screening goes wrong, because it is tempting to let software reject the majority of applicants automatically.
The safer approach
Let AI sort and summarize applications against the scorecard, with reasons for each placement. A recruiter reviews the bottom group before anyone receives a rejection. This keeps the speed and removes the risk of silent, unreviewed rejections.
Signals AI should not use
ZIP codes, graduation years, names, photos, gaps in employment and school prestige can all act as stand-ins for protected characteristics such as race and age. Illinois law now specifically bars using ZIP codes as a proxy for protected classes. Remove these signals from scoring.
Stage 3: Practical screening questions
Short questions about location, availability, pay expectations and work authorization save everyone time when they come early.
Where knockout questions go wrong
Hard knockouts reject anyone who answers outside a fixed range. A candidate who needs a slightly later start date or a hybrid arrangement may still be the best hire. Route near-misses to a recruiter instead of auto-rejecting, and never ask about salary history where local law bans it.
Stage 4: Outreach, follow-up and scheduling
Automate these fully. Candidates get faster replies, and recruiters get hours back each week. Keep the whole conversation in one thread per candidate that recruiters can see and take over at any point.
Keep outreach honest
AI-written outreach that is generic or gets details wrong damages your employer brand quickly. Personalize messages using the candidate's actual work, and let candidates know when they are talking with an AI assistant.
Stage 5: Skills assessments and structured AI interviews
AI-led structured interviews can give every candidate the same questions, at a time that suits them, with consistent scoring.
What AI should score
Score the content of answers against a rubric written for the role: technical accuracy, problem-solving approach, clarity of explanation and evidence of past results.
What AI should not score
Do not score facial expressions, tone of voice, accent, appearance or background surroundings. These measure nothing about job performance and can disadvantage candidates with disabilities or from different backgrounds. Pew found 70% of Americans oppose using AI to analyze workers' facial expressions. Illinois also requires notice, an explanation and consent before AI analyzes video interviews.
Pair AI interviews with a human follow-up
Treat the AI interview as the first round. Candidates who pass meet a person next, and that person reviews the AI's findings before the conversation.
Stage 6: Shortlisting and rejection
AI can propose a shortlist with the evidence for each candidate. The recruiter and hiring manager agree the final list.
Rejections stay with a person
Every rejection after a candidate has interviewed should be confirmed by a named person, with a recorded reason tied to the scorecard. A rejection message should be prompt and respectful, and candidates who reached an interview deserve more than a template.
Stage 7: Final interviews, references and the offer
Keep these entirely human. AI can prepare a briefing for the hiring manager with the full candidate record and open questions, but the conversation, the reference calls and the decision belong to people.
A note on executive hiring
For senior and executive roles, AI's job is smaller. It is useful for mapping the market, researching candidates' public track records and organizing interview notes. It should not screen executive candidates in or out. Leadership hires depend on judgment about context, relationships and stage fit that requires experienced people and careful references.
How to handle exceptions
A screening process is only as good as what happens when something does not fit. Decide the routing rules before the role opens, not when a hard case appears.
• AI confidence is low or scores conflict: flag the candidate instead of scoring them down. A recruiter handles it.
• Strong evidence outside the rubric, such as an unusual background or career change: send the candidate to manual review by a recruiter, then the hiring manager.
• A practical answer just outside the range: route the candidate to a recruiter instead of auto-rejecting.
• The candidate asks for a human: move the conversation to a recruiter within one business day.
• An accommodation request: skip all automated steps immediately. A recruiter or HR handles it.
• The candidate disputes a result: someone not involved in the original decision reviews it, usually HR or the hiring lead.
Make it easy to reach a person
Tell candidates in the first message how to ask for a human, and honor it without penalty. The option builds trust even when few people use it.
Keep a record of every screening decision
If you cannot explain why a candidate was advanced or rejected, you have a fairness problem and a legal one. Records also show you where your rubric or tools need work.
What to record for each decision
• The evidence used, with links to work samples or interview notes
• What the AI recommended, its score and its stated reasons
• Who made the final call, and when
• The reason for the decision, in plain words tied to the scorecard
• Any override of the AI recommendation, and why
How long to keep records
California's rules on automated decision systems require employers to keep AI-related employment records for four years (Washington University School of Law AI policy tracker). Many multi-state employers adopt the longest applicable period across their locations.
Tell candidates how AI is used
Candidates are more comfortable with AI when they know where it is used and that a person decides. Several laws now require notice, and it is good practice everywhere.
A sample candidate notice
We use AI tools to help us review applications and run a structured first-round interview. The AI looks only at job-related information, such as your work experience, work samples and answers to role-specific questions. It does not make hiring decisions. A member of our recruiting team reviews every result, and you can ask to speak with a person at any time by replying to this message.
Adapt this notice to your process and the laws where you hire.
US laws on AI candidate screening (2026)
The US has no single federal AI hiring law, but federal anti-discrimination laws, including Title VII, the Age Discrimination in Employment Act and the Americans with Disabilities Act, apply to automated screening just as they apply to people. A growing set of state and city laws adds specific duties. Dates change often; confirm current status before relying on this list.
• New York City, Local Law 144 (in force since July 2023): employers using automated employment decision tools need a bias audit within the year before use, a published summary and notice to candidates who live in the city (NYC DCWP).
• Illinois, HB 3773 (in force since January 1, 2026): amends the Illinois Human Rights Act to ban AI use with a discriminatory effect in hiring and to require notice when AI is used (WashU Law).
• Illinois, Artificial Intelligence Video Interview Act (in force since 2020): requires notice, an explanation and consent before AI analyzes video interviews.
• California, Civil Rights Council rules on automated decision systems (in force since October 1, 2025): apply anti-discrimination law to AI tools and require employers to keep records for four years (WashU Law).
• California, privacy rules on automated decision-making (in force since January 1, 2026): require a risk assessment before using automated decision-making for significant decisions about applicants (HR Executive).
• Colorado, Colorado AI Act as amended (scheduled for January 1, 2027 after delays): adds duties for employers using high-risk AI in hiring decisions (Ethena).
What this means in practice
A few habits cover most of these rules at once: keep people on the decisions, test your screening outcomes for bias, keep records, and tell candidates where AI is used. If you hire in New York City, Illinois, California or Colorado, check the specific duties with counsel.
How to set up AI candidate screening: step by step
These six steps work for a startup hiring its first engineers and for a larger team adding AI to an existing process.
1. Write the scorecard first. List the outcomes the role must deliver and the evidence that shows someone can deliver them. Everything AI scores must trace back to this.
2. Map each stage to a bucket. Use the framework above to mark every task as automate, assist or keep human.
3. Remove risky signals. Strip names, photos, ZIP codes, graduation years and school names from anything the AI scores.
4. Set exception rules and owners. Decide who handles low-confidence cases, near-misses, accommodation requests and disputes.
5. Tell candidates. Add a short notice to job posts and first messages.
6. Audit every month. Review a sample of AI recommendations against your team's decisions, and check outcomes across groups.
How to check for bias
Compare selection rates at each stage across demographic groups where you lawfully have the data. Under the long-standing four-fifths guideline used by US agencies, a selection rate for any group below 80% of the highest group's rate is a signal of possible adverse impact worth investigating.
How to measure whether AI screening is working
Track speed, quality and fairness together. Faster screening that produces worse hires or unfair outcomes is not an improvement.
• Time to qualified shortlist: shows whether automation saves real time. Watch for speed gains that come with lower shortlist quality.
• AI and recruiter agreement rate: shows whether AI recommendations match your team's judgment. Very low agreement suggests a weak rubric, and 100% agreement suggests people are rubber-stamping.
• Override rate and reasons: shows where the AI misjudges candidates. Watch for repeated overrides for the same type of candidate.
• Selection rate by group, by stage: shows whether any stage creates adverse impact. Investigate any group below four-fifths of the top rate.
• Candidate satisfaction: shows whether candidates trust the process. Watch for complaints about AI or unclear communication.
• Quality of hire at 90 days: shows whether screening predicts performance. Watch for hires who pass AI screening but struggle in the role.
How Talentz draws the line
Talentz.ai's recruiting platform, Zee, is built around the automate, assist, keep-human split. Each of its three agents handles the evidence work at one stage, and people make every hiring decision.
• Zee Scout (automate): finds candidates by proof of work on the platforms where that work lives. Recruiters review and approve the pipeline.
• Zee Engage (automate): runs two-way conversations across email, SMS, WhatsApp and voice on one thread per candidate, and confirms practical details. Recruiters can see and take over any conversation.
• Zee Lens (assist): runs structured AI video interviews and delivers findings with visible scoring reasons. Recruiters approve or reject, and the hiring manager makes the final call.
Why we built it this way
We believe AI should take on the work that buries recruiting teams, such as searching, following up and running consistent first interviews, and hand people better evidence for the decisions only they should make. That is also what candidates tell researchers they want: consistency from the process and a person accountable for the outcome.
See where humans stay in control. Start with one open role and review every Zee finding before anything reaches a candidate.
Frequently asked questions
What is AI candidate screening?
AI candidate screening is the use of software to find, review, question, assess or rank job candidates before a hiring decision. It can include sourcing, resume review, screening questions, scheduling, skills assessments and structured interviews. Good practice keeps people responsible for rejections and hiring decisions.
What parts of hiring should be automated?
Automate high-volume, repetitive work: sourcing, answering candidate questions, confirming practical details, outreach, follow-ups and scheduling. Use AI to assist with application review, structured interviews and scoring, with a person approving. Keep rejections, unusual cases, accommodations and final decisions with people.
Should AI reject candidates automatically?
No. AI can rank and flag candidates, but a named person should confirm rejections, especially after a candidate has applied for a specific role or interviewed. Unreviewed automated rejections create the highest risk of unfair outcomes and legal exposure.
Is AI candidate screening legal in the US?
Yes, but federal anti-discrimination laws apply to it, and some places add rules. New York City requires bias audits and candidate notice, Illinois bans discriminatory AI use and requires notice, and California applies its anti-discrimination rules to automated systems. Check the rules where you hire.
How do you use AI in hiring without bias?
Score only job-related evidence against a clear rubric, remove proxies such as ZIP codes, names and graduation years, keep people on decisions, and compare selection rates across groups at each stage. Investigate any group whose rate falls below four-fifths of the highest group's rate.
Do candidates accept AI in hiring?
Partly. Pew Research found 71% of US adults oppose AI making final hiring decisions, but 47% think AI would do better than humans at evaluating all applicants the same way. Candidates accept AI for consistency when a person stays accountable and the process is explained.
Can AI conduct job interviews?
AI can run structured first-round interviews, asking every candidate the same job-related questions and scoring answers against a rubric. It should not score facial expressions, tone or appearance. A person should review the findings, and candidates who pass should meet a person next.
What should you tell candidates about AI screening?
Tell them which steps use AI, what information it looks at, that a person makes the decisions, and how to reach a person. Some laws, including New York City's and Illinois's, require notice before AI is used.
Sources
• AI in hiring and evaluating workers: what Americans think, Pew Research Center, April 2023
• Gartner survey shows just 26% of job applicants trust AI will fairly evaluate them, Gartner, July 31, 2025
• Recruiters see job applications triple to more than 300 per role, HR Dive, May 2026
• Quarterly employment law update, Spring 2026, Gunderson Dettmer (Mobley v. Workday)
• Is cognitive ability the best predictor of job performance?, SIOP, on Sackett et al. (2022)
• Automated Employment Decision Tools FAQ, NYC Department of Consumer and Worker Protection
• AI policy and regulation resources: labor and employment, Washington University School of Law
• State vs. federal AI law: the map every HR executive needs, HR Executive
This article is general information, not legal advice. Laws on AI in hiring change often; confirm the rules that apply where you hire.