Key takeaways
- HeyMilo and Metaview lead the 34 AI candidate screening tools at 60/100, HeyMilo slightly ahead unrounded. They are the only two to reach a D; no tool in the category reaches a C.
- Screening tools screen very different things - resumes and applications, conversational pre-screens and AI interviews, applicants on a job board, or criminal records. Decide which kind you need before comparing scores.
- Seven vendors say a person makes every decision, or that they never auto-reject or auto-advance, while their own documentation describes moving or rejecting applicants automatically: HeyMilo, Braintrust AIR, Sense, Brainner, Zappyhire, Talkpush and Ribbon.
- Metaview and Fountain document real safeguards: on Metaview a screening rating does nothing until a recruiter acts, and Fountain's AI interview scores can only route applicants to a review stage. Checkr's background screening returns advisory values and automates the adverse-action notice process.
- Bias evidence is thin. Checkr, SEEK, Employment Hero and Workstream publish no independent bias audit; Metaview's and Wellfound's audits run on test data rather than real applicants.
AI candidate screening is the step where software reads an application before a person does: it scores the resume, asks the knockout questions, runs a short interview, or ranks the applicant list. It is also the step where a tool most easily decides who never reaches a recruiter.
We score every vendor in the directory on the same rubric, from public evidence. Below are the 34 tools filed under screening as their primary category, ranked by score, grouped by what they actually screen, with what each one documents about rejecting applicants automatically.
How we ranked them
Each vendor gets a 0-100 score in seven categories, weighted into one overall score:
| Category | Weight | What it asks |
|---|---|---|
| Article 11 technical documentation | 20% | Annex IV-style documentation of how the system works |
| Bias audit transparency | 18% | Published audits, methodology, auditor, cadence |
| FRIA support | 15% | Material a deployer needs for an impact assessment |
| Data governance disclosure | 15% | Training data, retention, subprocessors |
| Human oversight design | 12% | Review, override and audit-trail controls |
| Post-market monitoring | 12% | Change logs, incident handling, ongoing testing |
| Customer documentation | 8% | Deployer guidance, notices, instructions for use |
The score rates a vendor's public evidence across its AI hiring features. It is not a measure of how accurately a tool screens, and it is not an endorsement. Only two of the 34 reach a D; the rest are F.
AI screening tools at a glance
| Rank | Tool | What it screens | Score |
|---|---|---|---|
| 1 | HeyMilo | Voice, phone and video interviews | 60 (D) |
| 2 | Metaview | Applications, screening interviews | 60 (D) |
| 3 | Braintrust AIR | Voice interviews | 59 (F) |
| 4 | Fountain | Voice interviews, frontline applicants | 58 (F) |
| 5 | Sense | Chat, voice and video pre-screens | 57 (F) |
| 6 | Tenzo AI | Phone, text and video interviews | 56 (F) |
| 7 | Classet | Phone knockout interviews | 55 (F) |
| 8 | Checkr | Background-check records | 54 (F) |
| 9 | SEEK | Job-board applicants | 54 (F) |
| 10 | Wellfound | Startup applicants, resumes | 52 (F) |
| 11 | BrightHire | Voice screens, interview notes | 51 (F) |
| 12 | Handshake | Early-career applicants | 50 (F) |
| 13 | Stepstone | Job-board CVs | 49 (F) |
| 14 | Employment Hero | Applicants, AI interviews | 48 (F) |
| 15 | Pillar | Interview notes | 47 (F) |
| 16 | Crosschq | Resumes, interviews, references | 47 (F) |
| 17 | Brainner | Resumes in your ATS | 46 (F) |
| 18 | Talkpush | Chat flows, spoken English | 46 (F) |
| 19 | ConverzAI | Voice screens for staffing | 44 (F) |
| 20 | Ribbon | Voice and video interviews | 43 (F) |
| 21 | Tezi | Chat screens (withdrawn April 2026) | 43 (F) |
| 22 | Humanly | Chat, voice and video screens | 42 (F) |
| 23 | Tengai | Avatar interviews | 42 (F) |
| 24 | Workstream | Hourly applicants, voice and video | 39 (F) |
| 25 | Brazen | Hiring-event registrants | 38 (F) |
| 26 | HirePro | Resumes, AI interviews | 38 (F) |
| 27 | impress.ai | Resumes, chat pre-screens | 38 (F) |
| 28 | micro1 | Voice interviews, own marketplace | 36 (F) |
| 29 | Zappyhire | Resumes, chat, video | 36 (F) |
| 30 | Qureos | Applicants, voice and video | 33 (F) |
| 31 | Emi | Frontline chat screens | 33 (F) |
| 32 | Arena Analytics | Predicted retention | 32 (F) |
| 33 | XOR | Chatbot pre-screens | 30 (F) |
| 34 | Mercor | AI interviews, own marketplace | 28 (F) |
Tools sharing a rounded score are listed in unrounded order: HeyMilo ahead of Metaview at 60, Checkr ahead of SEEK at 54, Pillar ahead of Crosschq at 47, Brainner ahead of Talkpush at 46, Ribbon ahead of Tezi at 43, Humanly ahead of Tengai at 42, Brazen, HirePro and impress.ai at 38, micro1 ahead of Zappyhire at 36, and Qureos ahead of Emi at 33.
"A person decides" - except in the documentation
The question that matters most for a screening tool is whether it can reject an applicant without anyone looking. Seven vendors say a person makes the call, while their own pages say otherwise:
- HeyMilo's FAQ says it "never auto-advances a candidate", but its ATS write-back rules list "Auto-advance top scorers" and "Auto-route low scorers to a rejected pipeline".
- Braintrust AIR's product page says "AIR does not auto-reject candidates"; a help-centre article tells recruiters it lets them "Filter out unqualified candidates automatically".
- Sense's AI Policy says decisions such as hiring and rejection are made by people; its AI Interviewer page says candidates "can be dispositioned automatically".
- Brainner's homepage says advancing or rejecting "always requires human action"; its help centre documents rules that "automatically advance or reject candidates" and archive them "on autopilot".
- Zappyhire's ethics page says people oversee every hiring decision; its ZappyVue guide says candidates below the minimum score are rejected automatically, without notification.
- Talkpush's privacy policy says "no automated decisions are made by the platform", in the same document that describes optional filters which "automatically processes applicants according to matching criteria determined by the Employer".
- Ribbon's privacy policy says it does not make autonomous hiring decisions; its JazzHR page says customers can "auto-advance or disposition candidates" from interview results.
Others are plain about it. SEEK lets hirers auto-move applicants who miss a must-have to "Not Suitable" and emails them within 48 hours; Workstream's application form can reject applicants as soon as they submit; Humanly's knockout rules disqualify; Emi says its AI will "evaluate and filter candidates automatically".
A few document the opposite. On Metaview, a screening rating has no effect until a recruiter clicks Progress or Reject, and no question acts as a knockout. On Fountain, rules built on an AI interview stage cannot send anyone to Approved or Rejected, only to a review stage.
Resume and application screening
Metaview - 60/100
Metaview started as an AI interview notetaker and now sells Application Review, which sorts inbound applicants into match categories with written reasoning, plus AI screening interviews scored against rubrics recruiters write.
Strengths. The clearest no-auto-reject design in the category, a public subprocessor list and monthly product updates, and monthly Warden AI audits of Application Review and Screening.
Watch out for. The audits run on counterfactual test profiles, not applicant data, and its Sourcing and fillmore outreach agents have no independent audit. Its EU AI Act position covers only the notetaker; the approach for its screening features is still being prepared. Compare it with HeyMilo.
Brainner - 46/100
Brainner is a small AI screening layer for ATSs such as Workday, Lever and Greenhouse. Recruiters edit the criteria it drafts from the job description, and a language model scores and ranks applicants with per-criterion justifications tied to lines in the CV.
Watch out for. Its Idiro Analytics audit is available only on request, audit logs come only on the Enterprise plan, and its advice to always include candidate location as a filter or scored criterion deserves review against Illinois's ban on zip codes as a proxy.
Crosschq - 47/100, and the rest
Crosschq adds a five-star Resume Screen score, with bulk advance or disposition, to its reference checks and AI Interview Agent. Admins can switch off or tune the AI recommendation, but the only independent audit on record is Traitify's April 2025 audit, commissioned before Crosschq bought it and behind a login. impress.ai turns "must-haves into rules that triage large pools quickly" with no review language, and Zappyhire's help centre lists Gender as a selectable eligibility filter. Arena Analytics predicts who will stay in a role and "will only recommend people who scored as likely to be retained".
Job boards that screen their own applicants
SEEK - 54/100
SEEK runs job boards across Australia, New Zealand and Southeast Asia. Its seekAi labels applicants "high-fit" using a model trained on millions of real hirer shortlisting decisions, with an AI-written rationale.
Strengths. Strong documentation scores, an AI pre-screen that transcribes and summarises answers without scoring them, and membership of Singapore's AI Verify Foundation.
Watch out for. No published bias audit, although the model learns from past hirer decisions that can carry past bias, and the must-have filter's auto-reject path documents no review step.
Wellfound - 52/100
Wellfound, the startup marketplace formerly AngelList Talent, sells AI Resume Screening that writes must-have and nice-to-have criteria from the job description and ranks applicants.
Strengths. The second-highest bias audit score in the category: a monthly Warden AI audit linked from its own site footer, with impact ratios of 0.96 to 1.00, and a recruiter accepts or rejects each applicant.
Watch out for. The audit used Warden's own candidate dataset because historical data wasn't available, and covers only resume screening. Wellfound publishes no AI policy, model card or EU AI Act statement of its own.
Handshake, Stepstone and Employment Hero
Handshake ranks early-career applicants against employer criteria and rejects criteria it judges discriminatory, but its Holistic AI audit used synthetic CVs, passed its own 17 September 2026 expiry date and predates the July 2026 relaunch; criteria such as GPA cut-offs or graduation-year windows can produce adverse impact by race or age. Stepstone's Screening Agent marks each CV requirement as fully, partially or not met with reasons, but an optional Quality Check filters out non-EU applications automatically. Employment Hero moves applicants who meet every must-have forward automatically and leaves the rest for review, but publishes no independent audit and does not say what its interview "sentiment analysis" measures.
Conversational pre-screens and AI interviews
Many of the 34 run the screen as a conversation: a chatbot, a phone agent or an AI interviewer. Seven of them - HeyMilo, Braintrust AIR, Fountain, Sense, Tenzo AI, Classet and BrightHire - are ranked and compared in detail in our AI interviewers ranking, including what their bias audits tested. The short version: most of their audits run on test data, and Classet's is the only published audit in that group run on real candidates.
Further down, Talkpush grades spoken English as a pass/fail gate, and its validation study has no analysis by protected class; Workstream returns a pass/fail for each voice or video answer; XOR lets candidates "qualify themselves" against recruiter-set scores, with no bias audit; and Emi screens frontline applicants in Mexico, where at least one customer tells applicants decisions may be made without human intervention. Tezi withdrew its Max agent in April 2026. micro1 and Mercor mainly screen experts for their own marketplaces.
Background screening and interview notes
Checkr - 54/100
Checkr is a consumer reporting agency. Its AI classifies criminal charges across US jurisdictions, matches names and records, and checks documents for forgery; its Assess product applies employer rulesets to return advisory eligible, review or escalated values rather than final decisions.
Strengths. The joint-highest technical documentation score in the category, ISO 42001 certification (announced October 2025) and automated pre- and post-adverse-action notices with waiting periods and in-portal disputes.
Watch out for. No published bias audit of the charge classifier or Assess outputs, a claim that "the risk of bias is drastically reduced by 90%" with no published methodology, and Davis v. Checkr, filed in January 2026, which alleges records were matched to the wrong person. Compare it with SEEK.
Pillar, now Employ's AI Interview Companion, records and summarises interviews people run; its scores come from interviewers' own ratings, though it recommends next steps inside Lever.
The rules that apply to AI screening
- NYC Local Law 144: a screening score used as a substantial factor in a hiring decision makes the tool an automated employment decision tool, so the employer needs an annual independent bias audit, a published summary and candidate notice. An automatic rejection on a score is the plainest case: the score is the whole decision.
- Illinois HB 3773: notice when AI is used in employment decisions since 1 January 2026, and no zip code as a proxy for protected classes - relevant wherever location is a scored criterion.
- Colorado's SB 26-189: from 1 January 2027, notice, a plain-language explanation of an adverse decision and meaningful human review.
- EU AI Act: screening is high-risk under Annex III §4 when its output is decisive, with obligations from 2 December 2027; candidates need a way to contest.
How to choose an AI screening tool
- Decide what you are screening. Resumes, a conversation, a job board's applicant pool or a background check are different products with different risks. Pick the type first.
- Find every path to automatic rejection. Ask for the settings, not the policy page: knockout questions, score thresholds, ATS write-back rules and "Not Suitable" filters. Then decide where a person must sign off.
- Ask what the bias audit tested. Which feature, on what data, and when. Test profiles show whether demographic details change a score; they cannot show how your applicants fared. See what makes an LL 144 audit defensible.
- Check the criteria you can choose. GPA, graduation year, location and gender filters can each produce adverse impact.
- Ask for the documentation. Use our Annex IV walkthrough as the question list, and run the quick check for the EU AI Act.
For the wider market, see the best AI recruiting tools and the best pre-employment assessment tools. Every score above links to a full scorecard with the evidence behind each category, and any vendor can submit new evidence.
Frequently asked questions
What is AI candidate screening?
Software that reads, scores or ranks job applicants before a recruiter does. Some tools score resumes and applications against job criteria, some run conversational pre-screens or AI interviews, some are job boards ranking their own applicants, and some, like Checkr, classify background-check records.
What is the best AI candidate screening tool in 2026?
On HireAIScore's rubric, HeyMilo and Metaview score highest at 60/100, with HeyMilo slightly ahead unrounded. For resume and application screening, Metaview ranks highest. Scores rate each vendor's public evidence, not how accurately the tool screens, and no screening tool reaches a C.
Can AI screening tools reject candidates automatically?
Many can. SEEK, Workstream, Humanly, Emi and Zappyhire document automatic rejection or filtering, and seven vendors - HeyMilo, Braintrust AIR, Sense, Brainner, Zappyhire, Talkpush and Ribbon - say a person decides while their own documentation describes moving or rejecting applicants automatically. On Metaview a screening rating does nothing until a recruiter acts, and Fountain's AI interview scores can only route applicants to a review stage.
What laws apply to AI resume screening?
In New York City, a screening score used as a substantial factor makes the tool an automated employment decision tool under Local Law 144, which requires an annual independent bias audit and candidate notice. Illinois HB 3773 has required notice of AI use since 1 January 2026, Colorado's SB 26-189 adds notice, explanation and human-review duties from 1 January 2027, and the EU AI Act treats decisive screening as high-risk, with obligations from 2 December 2027.
How were these screening tools ranked?
Each vendor is scored on HireAIScore's rubric v1.0 from public evidence: Article 11 technical documentation (20%), bias audit transparency (18%), FRIA support (15%), data governance disclosure (15%), human oversight design (12%), post-market monitoring (12%) and customer documentation (8%). The ranking is the overall score across the vendor's AI hiring features.