Best AI candidate screening tools in 2026: resumes to first call

As of November 2026, the leading AI candidate screening tools are Eightfold, for large employers that want every applicant matched and scored against each job; HireVue, for enterprises that want a scored, structured screen backed by published explainability documents; Paradox, now part of Workday, for hourly hiring where candidates apply and are screened by text; Greenhouse Talent Matching and Ashby's AI-assisted application review, for in-house teams that want screening inside the ATS they already use; and Sense, for staffing agencies. Workable and Manatal are the lower-cost options with published prices, Humanly suits high-volume hourly roles, and Textkernel suits teams that want transparent, search-based matching inside an existing system. The products differ most in three things: whether they score or rank people, what the recruiter sees behind each result, and what bias audit the vendor publishes.

This page covers AI that screens, matches or ranks applicants before the first interview: resume review, knockout questions, job matching and text or chat pre-screens. AI that runs the whole interview is compared in best AI interviewer software, tests in pre-employment assessment tools, and AI voice screens in AI tools for phone screens. Screening is the step most AI hiring laws target, so read the legal section before you buy.

How we chose

• It screens people who applied. The product has to evaluate, score, label or sort inbound applicants before a human interview. Tools that only find new prospects are left out.
• The vendor says what it screens on. We included a product only if its own site or help center describes the inputs, the output and what a recruiter sees.
• Different buyers. The list covers large employers, in-house teams on a modern ATS, staffing agencies and high-volume hourly hiring.
• Facts from vendor sites, checked in November 2026. Every product fact links to the vendor's own page. Bias audit and compliance statements are the vendor's claims, attributed, not our findings. "Not published" means we could not find it on the vendor's site.

How do AI candidate screening tools compare?

As published by each vendor, as of November 2026. Sources are linked in each product section below.

What does the recruiter see, and what does each vendor publish on bias?

An explanation behind each result is what lets a recruiter catch a bad call, and a published audit is the first thing a regulator or a rejected candidate's lawyer will ask about. This is what each vendor says on its own site.

Which AI screening tools suit large employers?

Eightfold

Eightfold's matching model compares a job's titles, experience, education and skills with a candidate's work history, education and skills, and gives a match score from 0 to 5, in half-point steps, for each candidate and job; scores can sort candidates into a ranked list. Recruiters hover over a candidate to see how the profile matches the job's requirements, and Eightfold says the model does not replace human decision making (Eightfold matching model). It publishes a summary of a March 2026 BABL AI audit designed for New York City's Local Law 144, adding that no audit can show a model is bias-free (Eightfold bias audit results). Its own guides cover SAP SuccessFactors and Greenhouse; pricing is not published. Applicants have sued Eightfold under the Fair Credit Reporting Act; our AI hiring lawsuits page follows the docket. Best for: large employers that want every applicant scored against every relevant job.

HireVue

HireVue screens at two points. Match and Apply, an assistant on the careers site, reads a resume or asks about skills and interests, recommends jobs, and can ask qualifying questions synced from the ATS before someone applies (Match and Apply). The AI Interviewer is sold as the screening step: qualification questions return Meets or Does Not Meet, skill questions are scored on a 1 to 5 rubric, and the AI says why each answer does or does not meet the criteria; scores go to the ATS and recruiters decide (AI Interviewer). HireVue says its video assessments analyze only the transcript, not facial, video or audio data, and names ORCAA and Landers audits of its algorithms (HireVue AI in hiring). It also publishes an AI Explainability Statement and an integrations list that starts with Workday, SAP SuccessFactors and Oracle. Pricing is not published. Best for: enterprises that want a scored, documented screen for every applicant.

Paradox (now part of Workday)

Paradox's assistant, Olivia, screens applicants in a text conversation started from a QR code or a text-to-apply keyword. It checks each candidate against the job's requirements, schedules qualified people for interviews, and dispositions the rest or suggests other jobs (Paradox screening). The page describes no numeric score; in plain terms, it sorts people into qualified and not qualified. Paradox says gender, race and ethnicity never factor into screening and that it regularly runs bias evaluations, without naming an outside auditor (Paradox responsible AI); its integrations page has the ATS list. Workday completed its acquisition of Paradox on October 1, 2025 (Workday). Pricing is not published. Best for: hourly and frontline hiring where candidates apply from a phone.

Which ATSs have AI screening built in?

Greenhouse Talent Matching

Talent Matching, part of Greenhouse's Real Talent features, compares each resume with the skills a recruiter calibrates and labels candidates Strong, Good, Partial or Limited Match, showing highlighted resume terms, a summary, and matched and missing skills. It never advances or rejects anyone. Candidates who opt out, or who are in locations where AI use is restricted, are marked "Needs manual review" (Talent Matching guide). Greenhouse says WardenAI audits the algorithm monthly and at every change, that resume details are sent to OpenAI under a data processing agreement, that parsed resumes are cached for 30 days, and that personal data does not train models (data processing FAQ). The pricing page lists it from the Core plan up, with no prices published (Greenhouse pricing). Best for: in-house teams on Greenhouse that want match labels and a built-in opt-out.

Ashby AI-assisted application review

Ashby checks each resume against criteria you set for the job and returns Meets or Does not Meet for each one, with a rationale and citations to the resume; criteria it cannot decide are marked unknown. There is no overall score or rank: you filter by the criteria met, and the reviewer advances or rejects. It is opt-in, reviewers can override outputs, and Ashby supports applicant opt-out of AI processing (Ashby). A FairNow audit of the model, dated August 2024 and following Local Law 144's method, reports no group selected at less than 80% of the most favored group's rate (FairNow audit); ask for the latest report in Ashby's trust center. Foundations is $400 a month for up to 100 employees, with 1,500 AI credits a month; Plus and Enterprise are quoted (Ashby pricing). Best for: in-house teams that want evidence per criterion instead of one number.

Workable

The Workable Agent scores applicants' responses, CVs and profiles against up to 14 job criteria, including must-haves and disqualifiers you mark, and normalizes the result to an overall score from 0 to 100 that feeds a shortlist. Workable says every score has reasoning, every action is logged, recruiters can override any decision, protected characteristics are excluded by design, and the Agent is aligned with the EU AI Act; it describes no independent audit (Workable AI). Standard is $299 and Premier $599 a month for companies with 1 to 20 employees. Screening is an add-on billed at 1 AI credit per candidate, with 3,000 free credits per plan and 5,000 more for $600 (Workable pricing). Best for: small and midsize companies that want a priced, score-based screen.

Manatal

Manatal's AI recommendations pull the skills and requirements from your job description, score every candidate in your database against them and rank the strongest first, with a breakdown for each requirement and weights you can set for skills, experience or education (Manatal AI recommendations). Its help center lists skills, location, experience and education among the parameters and says AI outputs guide, not replace, your judgment (Manatal AI features); we found no bias audit for the feature. Professional is $15, Enterprise $35 and Enterprise Plus $55 per user per month billed annually, all with AI scoring (Manatal pricing). Check how location is weighted, since Illinois bars using zip codes as a proxy for protected classes. Best for: small agencies and HR teams on a budget.

Which screening tools suit staffing agencies and high-volume hiring?

Sense

Sense offers a chatbot with pre-screening, a pre-screen bot whose answers go straight into the ATS (Sense pre-screen), and candidate matching that scores applicants on experience, skills and preferences (candidate matching). Its Warden AI trust page says matching is audited monthly, most recently on October 14, 2026, with reports labeled for Local Law 144, the EU AI Act, Colorado's SB 26-189 and California's FEHA (Warden AI trust page for Sense). Sense's pages do not say what reasoning a recruiter sees behind a score. Starting prices are published: Staffing Teams from $2,000 a month, Talent Acquisition from $7,000 and AI Recruiter from $3,000 (Sense pricing). Best for: staffing agencies on Bullhorn or a similar ATS.

Humanly

Humanly screens by chat, SMS, email, voice, phone and video on availability, shifts, certifications, right to work, location and experience. Configurable rules advance qualified candidates and disqualify those who miss knockouts, every candidate is scored on the same criteria so the top 10 to 20 percent surface first, and results write back to the ATS (Humanly screening); the home page describes a ranked shortlist with transcripts behind each name (Humanly). Put plainly, it scores, ranks and can reject people automatically, so decide which knockouts it may act on alone. It lists "bias controls," but we found no published audit or price. Best for: high-volume hourly hiring.

Textkernel Source & Match

Source & Match turns a job posting into a search query, matches across your database and outside sources, and returns a ranked shortlist. Textkernel calls it "white box matching" because the criteria are visible and the query is editable, and results can be anonymized. There is a free trial with 500 credits, Professional from $99 a month and custom Enterprise pricing (Source & Match). Its responsible AI page says matching runs on a term-based search engine rather than AI, that users cannot search on gender or religion, that users remain the decision makers, and that an internal fairness checklist guides bias work; it names no independent audit (Textkernel responsible AI). Best for: agencies that want matching they can inspect inside an existing ATS.

How do you choose an AI screening tool?

• Start with your ATS. On Greenhouse, Ashby, Workable or Manatal, try the built-in feature first: no new vendor, no sync to break, and the opt-out and audit trail live in the same system.
• Large employer, many requisitions: Eightfold or HireVue, both with published audit material and long integration lists. Budget for a sales process, since neither publishes prices.
• Hourly and frontline volume: Paradox or Humanly, where candidates apply and are screened by text within minutes. Our guide to AI tools for high-volume hiring covers the rest of that funnel.
• Staffing agency: Sense for Bullhorn-style desks, Textkernel if you want matching you can inspect, Manatal if budget leads.
• Choose by output, not only features. A per-criterion verdict (Ashby) or a match band (Greenhouse) is easier to review and explain than a single number. A 0 to 100 or 0 to 5 score is easier to sort but harder to defend when a rejected candidate asks why.

What should AI do at the screening step, and what must a human still do?

AI is good at reading every application against the same criteria, asking the same knockout questions at 9 p.m., and pointing to the line in a resume behind a result. It should not be the only thing between an applicant and a rejection. A human still has to:

• Write the criteria. Every tool here screens against requirements you set, so vague or inflated must-haves become automated rejections at scale.
• Review rejections, at least by sample. Each week, read some of the people the tool screened out or ranked low, and every knockout rejection where the question could be read two ways.
• Read the reason, not just the score. If a tool cannot show why it ranked someone low, treat the output as a hint.
• Handle accommodations and opt-outs. A candidate who asks for another route needs a real one: a recruiter review, more time or a different format. Greenhouse, Ashby and HireVue describe opt-out routes; for the others, ask.
• Watch adverse impact. Compare pass rates by sex and race or ethnicity at least quarterly; a group passing at under four-fifths of the top group's rate is the usual first warning sign.
• Decide in a human interview. The screen narrows the list; the first interview is where a person judges the candidate.

For that first human interview, Interview Signal, a desktop app, gives the interviewer a live question guide and a scorecard in which each score carries the candidate's own quotes; it does not screen resumes. We make Interview Signal; everything about the other products on this page comes from their own sites, linked and dated.

Which laws apply to AI candidate screening?

Tools that score, rank or filter applicants are the main target of AI hiring rules. Briefly, with the primary source for each, checked in November 2026. This is not legal advice.

• New York City, Local Law 144. DCWP has enforced the law since July 5, 2023 (NYC DCWP). Its rules require a bias audit no more than a year before use, a public summary of the results, and notice at least 10 business days before use that explains how to request an alternative selection process or a reasonable accommodation, plus published information on the tool's data retention policy and data sources. The rule's own example says a tool used only to screen resumes early in the process still needs an audit (DCWP final rule). Our Local Law 144 bias audit guide explains what the audit must contain.
• Illinois, HB 3773. Since January 1, 2026, Public Act 103-0804 has made it a civil rights violation for an employer to use AI in recruitment or hiring that has the effect of discriminating on a protected basis, to use zip codes as a proxy for a protected class, or to fail to give notice that it uses AI (Public Act 103-0804). See Illinois HB 3773 for the status of the notice rules.
• California. Civil Rights Council regulations under the Fair Employment and Housing Act, in effect since October 1, 2025, make clear that an automated-decision system can violate the Act if it discriminates on a protected basis, and require keeping automated-decision data for at least four years (California Civil Rights Department). See California's AI hiring regulations.
• Colorado. The 2024 Colorado AI Act was delayed and then rewritten. SB 26-189, signed May 14, 2026, repealed and reenacted it as rules on automated decision-making technology; it takes effect January 1, 2027 and applies to consequential decisions, including employment, made on or after that date. Employers using a covered tool owe candidates notice, a plain-language explanation within 30 days of an adverse outcome, and a way to correct data and ask for meaningful human review (Colorado General Assembly). Tools that only organize or present information for human review, without a score, ranking or recommendation that materially influences the decision, are outside it. See Colorado's AI law for employers.
• EU AI Act. AI intended to analyze and filter job applications and evaluate candidates is high-risk under Annex III, and the European Commission says the high-risk rules for these areas apply from 2 December 2027 (European Commission). See the EU AI Act and recruitment.
• Title VII disparate impact. A neutral practice, including an AI screen, that excludes a protected group at a disproportionate rate must be job related and consistent with business necessity (42 U.S.C. § 2000e-2(k)). The EEOC's 2023 technical assistance on AI selection tools was removed from its site in January 2025, so do not rely on it; the statute still applies. In Mobley v. Workday (N.D. Cal. No. 3:23-cv-00770), the court's July 12, 2024 order dismissed the intentional discrimination claims but let the disparate impact claims go forward on the theory that a screening vendor can be liable as the employers' agent (order); for where the case stands now, see AI hiring lawsuits. More in Title VII disparate impact and AI hiring, and every state in AI hiring laws by state.

Questions to ask a vendor before you buy

1. Bias audit. Who performed the latest audit, when, on which model version, and with whose data? Under New York City's rules you can rely on an audit built from other employers' data only if you supplied your own data to it or have never used the tool. Ask for the full report, not only the summary.
2. Adverse-impact monitoring. Does the tool report pass rates by group for your own requisitions, how often, and can you export them? Where a vendor says audits run monthly, ask whether that covers your data or a pooled sample.
3. Data retention. How long are resumes, parsed data, scores and explanations kept, and can you set it? California expects automated-decision data kept four years, and New York City requires you to publish the tool's retention policy. Ask what is sent to third-party AI models and whether it trains them.
4. Candidate notice. Does the vendor supply notice wording, and can it go in the job post or on the careers page 10 business days ahead for New York City roles? Illinois and, from 2027, Colorado also require notice.
5. Opt-out and accommodation. What happens to a candidate who opts out or asks for an accommodation: who reviews them, and does the tool flag them so they are not silently dropped?
6. Automatic rejection. Can the tool reject on its own, and can you switch that off per question or per job?
7. Explanations. Can a recruiter see, and export, the reason for each result if a candidate or regulator asks?
8. Model changes. Will you be told when the model changes? From January 1, 2027, Colorado requires developers of covered tools to give deployers documentation and notice of material updates.

A short setup checklist

1. Pick one requisition with plenty of applicants and clear, job-related must-haves.
2. Write the criteria and knockout questions, and remove anything that is not job related, such as zip codes.
3. Publish the candidate notice and the opt-out or accommodation route before the tool runs, 10 business days ahead for New York City roles.
4. Turn off automatic rejection for the pilot, and have a recruiter review the lowest-ranked and screened-out applicants.
5. Run the tool for two to four weeks beside your current process, and read every case where the tool and recruiters disagree.
6. Check pass rates by group before widening use, and keep the audit, notice and review records together.