AI screening tools that explain rejections: adjustable criteria and the reason behind every result

As of November 2026, the AI applicant screening tools that let you edit the criteria and see the reason behind each result are Workable's Agent (an editable ideal candidate profile with must-haves, nice-to-haves and disqualifiers, and reasoning for every criterion), Ashby's AI-assisted application review (up to 50 criteria per job, each marked meets, does not meet or undecided with its reasoning), and Greenhouse Talent Matching (a weighted calibration, with matched and missing skills and a short justification, kept on the candidate once the application is rejected). Manatal lets you weight criteria and shows a breakdown per requirement. Only some of these reject anyone on their own: Workable's auto-disqualify is off by default, and Greenhouse and Ashby leave every rejection to a person.

We make Interview Signal, which is mentioned once below for the interview stage; everything about the other products comes from their own sites and help centers, linked and dated. For the wider market, including bias audits and prices for ten screening tools, see the best AI candidate screening tools; this page goes deeper on two things only: editing the criteria and reading the reasons.

Which AI screening tools let you adjust the criteria and see why someone was rejected?

What counts as a "reason" in these tools?

Three different things travel under the word, and only one of them can be checked quickly:

• A verdict per criterion with reasoning. Workable, Ashby and Greenhouse say which criterion was met or missed and why. A recruiter can open the resume and confirm or overturn it. Greenhouse adds highlighted matched terms, so the evidence is on the page.
• A score with a breakdown. Manatal shows where a candidate "fits or misses the mark" per requirement. Useful for ranking; check that the breakdown names evidence, not just a number.
• A rule outcome. Knockout rules, such as Humanly's, explain themselves only as well as the question did: "did not meet knockout criteria" is a reason if the criterion is a real requirement.

Then there is the reason a person records when they reject someone, which most ATSs keep as a disposition or disqualification reason. Workable's Agent writes its auto-disqualifications into "your account's own disqualification reasons," and Greenhouse says it will "retain the match results and reasoning on the candidate the moment their application is rejected," available in the candidate packet until anonymized or deleted. Keep the two apart: the AI's reasoning is evidence; the recorded rejection reason is your decision.

How do you adjust ideal-candidate criteria without creating a problem?

1. Start from the job, not from past hires. Write criteria from the job description and the scorecard: skills, licenses, experience that the work needs. Criteria copied from "people like our best hire" import whatever those people had in common.
2. Let the tool's guardrails speak. Ashby scans criteria prompts for EEO problems and flags them; Greenhouse blocks protected attributes such as gender and warns when a skill could act as a proxy for one. Treat a warning as a stop, not a suggestion.
3. Separate must-haves from disqualifiers. A disqualifier should be a requirement you would enforce by hand for every applicant, such as a required license. Anything softer belongs in must-have or nice-to-have, where it moves a score instead of ending an application. Work authorization questions have lawful wording; see work authorization questions in interviews.
4. Re-run after every edit, and know the cost. Ashby offers "Save and Re-Evaluate all active Job Considerations"; Workable evaluates new and in-flight candidates against the edited profile with "no AI credits consumed," marks candidates scored under the old version with a banner, and charges one AI credit to re-evaluate each of them.
5. Version the criteria. Record what changed, when and who changed it. When someone asks why a candidate was rejected in March, the answer depends on the criteria in force in March.

How do you check that the explanations are right?

Before you trust a tool's reasons, test them on candidates you already know:

1. Take a sample of recent applicants for one role, for example 20 the tool would reject and 20 it would advance (example numbers), and read each explanation against the resume.
2. Count the explanations that cite evidence that is not there, or miss evidence that is. Ashby lets reviewers "Flag Evaluation" and Workable lets you "Edit candidate's fit"; use them, and track how often you have to.
3. Look hard at "unknown," "undecided" and "needs manual review" results; they should reach a person, not fall out of the pipeline.
4. Check non-traditional resumes: career changers, gaps, military experience, foreign degrees. They are where keyword-style matching fails first.
5. Compare selection rates across groups where you lawfully have the data. The four-fifths rule of thumb is still printed in 29 CFR part 1607, but as of November 2026 the Justice Department considers the Uniform Guidelines unlawful insofar as they impose disparate-impact liability without intent, and OPM and OFCCP have removed their references to them (91 FR 48234); Title VII's disparate-impact provision, state laws and New York City's Local Law 144 bias audits still apply.

Do candidates have a right to an explanation?

Sometimes, and the rules differ by place. Not legal advice; the AI hiring laws by state page covers each one.

• EU and UK candidates. The GDPR gives a right not to be subject to a decision "based solely on automated processing" with significant effects (Article 22), and a right to "meaningful information about the logic involved" where such decisions are made (Article 15(1)(h)).
• Colorado. SB 26-189, signed May 14, 2026, requires a deployer to give a plain-language description of a covered tool's role within 30 days after a consequential decision with an adverse outcome, and lets the person ask for meaningful human review; the attorney general must adopt rules on those disclosures by January 1, 2027.
• New York City. Local Law 144 requires a bias audit and advance notice to candidates, not an individual explanation.

Whatever applies, a tool that keeps per-criterion reasoning on the record makes the answer possible; a tool that only keeps a score does not.

What about explaining decisions after the interview?

Screening tools explain why an application moved or stopped. The same question comes back after interviews, where the evidence is what the candidate said. Interview Signal, the app we make, is not an application screener; it scores live interviews, and each score in its scorecard carries the candidate's own quotes, checked against the transcript (a quote not in the transcript is removed). It has no ATS integration, so the scorecard is pasted into the ATS by hand; how the verified-quote scorecard works shows the output. For the resume stage done by hand, see how to screen resumes faster.

Questions to ask a vendor before you switch on AI screening

• Can we edit every criterion, and does editing re-score candidates already in the pipeline? At what cost?
• For a rejected candidate, what exactly is kept: the verdict per criterion, the reasoning, the evidence, the criteria version? For how long?
• Can the tool reject anyone on its own? Is that on or off by default, and can we limit it to true knockouts?
• What happens to candidates who opt out of AI processing? Greenhouse labels them "Needs manual review"; Ashby skips the evaluation and records why.
• Can we export the reasons for an audit or a candidate request without screenshots?