AI Features — How They Work

This page describes, in plain language, what each AI feature in Zerva evaluates and how it produces its output. It is published under Section 7.2 of the Master Service Agreement so that hiring offices using Zerva can prepare any applicant notices required by automated-employment-decision laws. Hiring offices remain responsible for the accuracy and sufficiency of their own notices.

All AI output in Zerva is advisory. It is a starting point for a human recruiter's judgment, not a decision. Under our agreement, hiring offices may not use it as the sole or determinative basis for any employment decision.

AI provider. AI features run on models provided by Anthropic, under commercial terms that prohibit training on customer or applicant data. Zerva does not train AI models on applicant data.

Résumé and application scoring

• Input: the text of a résumé or application form responses the applicant submitted, plus the job title and any screening criteria the hiring office configured.
• What it evaluates: how the applicant's stated experience, skills, and responses relate to the hiring office's configured criteria for the role.
• Output: a short written summary and an advisory score or ranking, shown to the recruiter alongside the full original résumé or application.
• What it does not do: it does not access any information outside what the applicant submitted; it does not use photos, names, addresses, or demographic information as scoring criteria; it does not make or recommend a hiring decision.

Interview summaries and scoring (recorded interviews)

• Input: a text transcript of a recorded live interview, or an applicant's one-way video submission, where the hiring office has enabled recording and obtained any required consent.
• What it evaluates: the content of what was said — the applicant's answers as transcribed — against the interview questions and any criteria the hiring office configured.
• Output: a written summary of the conversation and, where enabled, an advisory score, shown to the recruiter alongside the recording and transcript.
• What it does not do: it does not perform facial recognition, facial-expression or emotion analysis, voice or tone analysis, or voiceprint identification, and it does not create biometric identifiers. Speaker names in transcripts come from the meeting platform's participant list, not from analysis of anyone's voice.

Bulk résumé import

• Input: PDF résumés uploaded by the hiring office.
• What it does: extracts contact details and work history from the document into structured applicant records so the recruiter does not have to type them in.
• Output: an applicant record the recruiter can review and correct. No scoring is applied at import.

Applicant rights

Applicants who have questions about how their information was evaluated, or who wish to request deletion of interview recordings, should contact the hiring office they applied to. The hiring office controls that data; Zerva processes it on their behalf.

Divinify LLC · info@zerva.us · zerva.us/legal