AI recruiting glossary: plain-English definitions of the terms you will hear

This AI recruiting glossary defines, in one sentence each, the terms recruiters and hiring managers meet when they buy, use or are asked about AI in hiring: tool types such as AI note takers, interview intelligence and AI interviewers; interview terms such as structured interviews and anchored scales; and legal and data terms such as adverse impact, AEDT, bias audit and data processing agreement. Terms are in alphabetical order, and each links to a fuller guide where one exists. Definitions reflect usage as of October 2026.

Legal terms are described in general terms only. Where a law defines a word precisely, the linked guide quotes the statute. Nothing here is legal advice.

Terms from A to B

• Adverse impact: a substantially lower rate of selection for a race, sex or ethnic group at a hiring step, which federal guidelines generally flag when a group's rate falls below four-fifths of the highest group's rate (the four-fifths rule, worked through).
• AEDT (automated employment decision tool): New York City's term for software built on machine learning, statistics, data analytics or AI that issues a simplified output, such as a score or recommendation, used to substantially assist or replace hiring or promotion decisions (Local Law 144 and note takers).
• AI agent: software that uses an AI model to carry out a multi-step task on its own, such as finding candidates, sending outreach and booking a call, rather than answering one prompt at a time.
• AI disclosure: a notice telling candidates which AI tools are used in hiring, what they do, and whom to contact for questions or an alternative (disclosure templates).
• AI-generated answer: an interview answer produced wholly or partly by an AI tool, whether rehearsed beforehand or fed to the candidate live; most employer policies allow the first and ban the second (candidate AI policy).
• AI interviewer: software that conducts an interview itself, asking questions by voice, video or chat and often scoring the answers, as opposed to a tool that records a human-led interview.
• AI note taker: software that captures a call's audio, transcribes it and turns it into notes or summaries; in recruiting it is used mostly on screens and interviews (how recruiters use them).
• Applicant tracking system (ATS): the system of record for requisitions, applications, candidate notes and hiring stages (writing notes in your ATS).
• ATS integration: a connection that lets another tool create or update ATS records automatically, instead of a recruiter copying and pasting (how to test one).
• Automated decision-making technology (ADMT): a term used in some state laws, including Colorado's, for technology that processes personal data to make or materially influence decisions about people, including hiring decisions (Colorado's law for employers).
• Behaviorally anchored rating scale (BARS): a rating scale in which each score is described by observable behavior, so every interviewer rates against the same standard (anchor library).
• Bias audit: an independent statistical check of whether a hiring tool's outcomes differ by sex, race or ethnicity, which New York City's Local Law 144 requires within the year before a covered tool is used (what the audit contains).
• Biometric identifier: a measurement of the body, such as a voiceprint, fingerprint or face geometry, which Illinois' Biometric Information Privacy Act regulates when hiring tools collect it (BIPA and hiring tools).
• Bot (meeting bot): an AI note taker that joins a video meeting as a visible participant in order to record and transcribe it (note takers without a bot).
• Bot-free capture: recording or transcribing a call on the interviewer's own computer or device, so nothing joins the meeting as a participant (recording Zoom without a bot).

Terms from C to D

• Calibration: a session in which interviewers score the same answer separately and then compare, so they apply the rating scale the same way (running a calibration session).
• Candidate fraud: deception in the hiring process, such as a proxy interviewer, deepfaked video or audio, or a fabricated identity or work history (spotting identity fraud).
• Candidate matching: AI that compares a resume or profile with a job's requirements and suggests or ranks likely fits, which is evaluation even when the vendor calls it a suggestion (sorting AI uses by risk).
• Comparison matrix: a table that puts candidates for one role side by side against the same criteria and evidence (matrix templates).
• Consent (recording consent): agreement from the people on a call to be recorded or transcribed; some US states require every party's consent and others only one party's (two-party consent states).
• Data processing agreement (DPA): the contract setting out how a vendor handles personal data on your behalf, including security, sub-processors, model training and deletion (what to check in a DPA).
• Deepfake: AI-generated or AI-altered video or audio that makes a person appear to say or do something they did not, including someone impersonating a candidate on a call.
• Deployer: under the EU AI Act, the organization using an AI system under its authority, such as an employer using a vendor's screening tool, as distinct from the provider that builds it (EU AI Act and recruitment).
• Diarization (speaker diarization): the part of transcription that works out who spoke when, so each line of a transcript is attributed to the right person.
• Disparate impact: the legal theory under which a neutral-looking practice, including an AI tool, can be unlawful if it disproportionately excludes a protected group and is not job-related and consistent with business necessity (Title VII and AI hiring).

Terms from E to H

• EU AI Act: the European Union's regulation on artificial intelligence, which lists AI systems used for recruitment and selection among its high-risk uses.
• Evidence-based feedback: interview feedback that records what the candidate said and did separately from the interviewer's interpretation and rating (the method).
• Four-fifths rule: the federal rule of thumb that a group's selection rate below 80% of the highest group's rate is generally regarded as evidence of adverse impact.
• Generative AI: AI that produces new text, audio, images or code in response to a prompt, such as drafting a job ad or summarizing an interview.
• Hallucination: AI output that is fluent and confident but false, such as a quote the candidate never said or an employer that is not on their resume.
• High-risk AI system: the EU AI Act's category for uses with the strictest obligations, which includes AI used to recruit, select or evaluate candidates.
• Human in the loop: a process in which a person reviews AI output and makes the decision, with the authority, information and time to overrule the tool.

Terms from I to L

• Impact ratio: in a bias audit, the selection rate (or scoring rate) of one group divided by that of the most-selected group.
• Interview guide: the planned questions for an interview, mapped to the competencies being assessed, with follow-ups and timing (interview guide template).
• Interview intelligence: software that records, transcribes and analyzes interviews to produce notes, scorecards, interviewer coaching or hiring data (what it is and how to choose).
• Knockout question: a screening question with a required answer, such as a license or shift availability, that removes applicants who do not meet it (pre-screening questionnaire).
• Large language model (LLM): the kind of AI model behind most generative AI tools, trained on large amounts of text to understand and produce language.
• Live transcript: a transcript produced while the call is still happening, so the interviewer can read back and a tool can suggest follow-ups in real time (transcription software for recruiters).

Terms from M to R

• Must-have (gate): a requirement a candidate has to meet regardless of strength elsewhere, checked before any weighted total (comparing candidates).
• Prompt: the instruction given to a generative AI tool, including any text pasted in for it to work from (prompts for building a scorecard).
• Provider: under the EU AI Act, the organization that develops an AI system and places it on the market, which carries most of the design and documentation duties.
• Quality of hire: a measure of how well new hires perform and stay, usually built from performance ratings, retention and hiring manager satisfaction (measuring quality of hire).
• Resume parsing: software that extracts fields such as employers, dates and skills from a resume into structured data in an ATS.
• Retention period: how long candidate data such as transcripts, notes and scorecards is kept before it is deleted (how long to keep interview notes).
• ROI (return on investment): the value a tool produces minus its cost, divided by its cost (AI note taker ROI, worked).

Terms from S to W

• Scorecard: a structured form listing the competencies to assess, the rating scale and space for evidence, completed by each interviewer (scorecard template).
• Selection rate: the share of candidates in a group who pass a hiring step, the input to adverse impact and impact ratio calculations.
• Skills-based hiring: assessing specific skills directly, through questions, work samples or tests, instead of relying on degrees or job titles as proxies.
• SOC 2: an independent auditor's report on a service provider's security and data controls, often requested from recruiting software vendors (SOC 2 for recruiting software).
• Speech-to-text (automatic speech recognition): technology that converts spoken audio into written text, the first step in any AI note taker (transcribing an interview).
• Structured interview: an interview in which every candidate for a role is asked the same job-related questions and rated on the same anchored scale (running one without an ATS).
• Submittal: an agency recruiter's write-up presenting a candidate to a client, with evidence against the brief, pay and availability (submittal template).
• Summarization: AI condensing a transcript or document into a shorter account, which can drop, merge or overstate details and so needs checking against the source.
• Training data: data used to build or improve an AI model; contracts for recruiting tools should say whether candidate data may be used this way.
• Transcript: a written record of what was said on a call, ideally with each line attributed to a speaker.
• Weighted scorecard: a scorecard in which competencies count for different shares of the total, with the weights set before interviews begin (weighted scorecard).
• Word error rate (WER): a standard measure of transcription accuracy, calculated as the substitutions, deletions and insertions needed to match a reference transcript divided by the number of words in that reference.

Which terms do people mix up most often?

A few pairs cause most of the confusion in vendor calls and policy drafts.

For how these terms fit together in practice, start with how to use AI in recruiting responsibly. For the rules behind the legal terms, see AI hiring laws by state, and for the numbers people quote about all of this, the AI in recruiting statistics page links every figure to its original source.