The words a hiring decision gets argued in.
53 terms from AI hiring, selection science and the statutes that govern both. Defined in plain language and linked to the pages that go deeper. Written for anyone who has to explain a screening process to a hiring manager, a candidate or a lawyer.
30-day free trial with 75 Hire Credits. Card required, nothing charged for 30 days.
- Scoring and evidence
- Fairness and the law
- Integrity and proctoring
Three different kinds of word live on this page.
AI hiring vocabulary mixes regulation, research and product naming. Knowing which kind of term you are looking at changes how much weight it carries.
Some of these are legal terms.
Adverse impact, bias audit and Automated Employment Decision Tool appear in statutes. Use them loosely in a vendor call and you can end up with obligations you did not price in. Where a term has a legal meaning, this page gives the legal meaning first.
Some are product terms.
Reportcard, round type, Hire Credits and readiness score describe how Expert Hire is built. They are defined here as they behave in the product. If a definition names a number or a limit, it is the number in the product today.
Some are science, not software.
Structured interview and predictive validity come out of decades of selection research, long before any of this was automated. The method was already better. It was just too expensive to enforce by hand.
Where a claim is honest, it is stated.
No tool removes bias. What software can do is remove identifying details before a review, hold every candidate to the same rubric, keep the evidence attached to the score, and leave the decision with a person. That framing runs through every definition here.
Grouped by where the word gets used.
Six groups, alphabetical inside each one. Jump to a group, or use the A to Z strip to land on the first term under a letter.
Interview formats and rounds
The shapes an evaluation can take. Getting these names right matters, because a screening call, a structured round and an assessment are three different things with three different levels of defensibility.
AI interview
A live interview conducted by an AI interviewer rather than a person, held over voice and scored against the same rubric as a human round. Expert Hire's AI interviewer is named Ethan; it asks follow-up questions based on what the candidate actually said and grades communication, confidence and technical depth. Because it runs 24/7 in the candidate's own language, first rounds stop being limited by recruiter calendars.
Async video interview
A screening format where a candidate records answers to pre-set questions on their own time and a reviewer watches later. It removes the scheduling problem but not the evaluation problem, since a recording is only as useful as the rubric applied to it. A scored AI round covers the same scheduling gap and returns a structured result instead of a video queue.
Coding test
An assessment where a candidate writes code that is really executed against hidden test cases, rather than compared to a stored answer. Expert Hire runs coding tests in a sandbox. Each test case comes back pass or fail with its runtime in milliseconds, plus a complexity verdict. Scores cover correctness, efficiency and code quality. It is one of the five round modes and lands on the same rubric as the interview rounds.
Human-led interview
A live round run by a recruiter or hiring manager, hosted inside Expert Hire or on Zoom, Meet or Teams with a notetaker in the room. It is recorded, transcribed and scored on the same rubric and the same reportcard as an AI round, so a human round stays comparable to an automated one. A recruiter can also silent-listen to a live AI interview and take over in real time.
Mock interview
A practice interview run for the candidate's own benefit rather than an employer's, targeted at a specific company, role, round type and difficulty. Attempts are repeatable and stored with a score, the change since the previous attempt and a short insight line naming what to work on next. Practice runs on the same engine as a hiring round, which is why the score means something.
Prompt assessment
A round that measures AI fluency by asking a candidate to write prompts for real-world tasks, scored on 8 dimensions including prompt quality and iteration strategy. Assessments are catalogued by category, by technical or non-technical tag and by level, with a set number of attempts per question. It answers a question a coding test cannot: can this person get useful work out of an AI tool.
Resume screening
The first-pass filter that reads a resume against a role and produces a fit score without an interview. Expert Hire runs it as its own round type, analysing skills, experience and education. You can bulk upload, sort the candidate table, shortlist the top of a list and export to CSV. It costs 1 Hire Credit per scan, which makes it the cheapest way to compress a large applicant pool.
Round type
Two levels sit under this word. A round has one of five modes: Resume Screening, AI Interview, Human-Led Interview, AI Prompt Assessment or Coding Test. Inside an AI interview you then pick one of eight round types: Coding, General Interview, Communication, HR Round, Problem Solving or Aptitude, System Design, Academic or Task. A pipeline stacks rounds, AI and human rounds mix freely, and all of them score onto one rubric.
Structured interview
An interview where every candidate is asked the same questions and scored against the same predefined criteria. Decades of selection research find structured interviews more predictive of job performance than unstructured conversation, and easier to defend if a decision is challenged. The hard part is enforcement, because a busy panel drifts. A scripted round and a shared rubric fix that.
Scoring, evidence and reports
How a conversation becomes a number, and what has to sit underneath that number for it to survive a debrief.
Readiness score
One comparable score that every round a person takes rolls into, rather than a separate result per stage. For an employer it is a per-candidate score built from interviews, resumes and coding on one rubric. For a student it is a personal readiness level. For a campus it aggregates into a batch readiness score. Two candidates assessed in different weeks by different rounds still land on the same scale.
Reportcard
The share-ready report produced for a candidate after a round. It carries an overall score out of 100, a verdict line and a fit chip. Under that sit the skill breakdown, the recording, a searchable speaker-separated transcript and the malpractice status. Tabs cover Overview, Analysis, Submissions and Resume, and the report can be shared or downloaded as a PDF. Reports are always included and never metered.
Rubric
The fixed set of criteria a round is scored against, written before anyone is interviewed. In Expert Hire you pick the skills for the job and set a weightage on each one, and you set up the job description. Answers are scored against both. All five round modes score onto the same rubric, which is what makes a resume screen and a coding test comparable inside one pipeline.
Explainable AI (XAI)
AI whose output a human can inspect and account for, rather than a number with no visible derivation. If you cannot say why a candidate scored what they scored, you cannot defend the decision. Every Expert Hire score traces back to the transcript, the live code, the skills you weighted and the JD you set up.
Human-in-the-loop (HITL)
A design where a person, not the model, makes the consequential decision. Expert Hire surfaces signals, scores and integrity flags to a human reviewer alongside the evidence; a flag is never an automatic verdict and we do not auto-reject anyone. The AI compresses the reading, the human still does the deciding.
Industrial-organizational (I-O) psychology
The scientific study of behaviour at work, and the field that produced most of what is known about which selection methods actually predict performance. Structured interviews, work samples and validated rubrics all come from this literature. It is the reason a coding test that runs real code beats a conversation about code.
Predictive validity
How well a selection method actually predicts later job performance, measured after the fact rather than asserted up front. It is the only real test of an assessment: a round can feel rigorous and still predict nothing. Structured, job-relevant rounds consistently outperform unstructured ones on this measure, which is the whole argument for scripting an interview.
Sentiment analysis
Reading emotional tone from language, usually to estimate confidence, enthusiasm or stress. It is a weak signal on its own and a poor basis for a hiring decision, because tone varies with culture, nerves and audio quality. Expert Hire grades communication and confidence as named dimensions inside a structured rubric with the transcript attached, rather than as a standalone mood score.
Transcript
The written record of what was actually said in a round, speaker-separated and searchable, with filters for the candidate and the interviewer. It is the primary evidence behind every score on a reportcard, and it is what a debrief should be argued from. Without it, an AI score is an assertion.
Fairness, law and audit
The vocabulary regulators use. If you run automated screening in New York City, Illinois, Colorado or the EU, these are the words that appear in the statute rather than in a brochure.
Adverse impact
A selection process that passes one group at a substantially lower rate than another, even when the process was not designed to discriminate. It is measured from outcomes, not from intent, which is why a tool can be neutral on its face and still produce a problem. Monitoring it means comparing selection rates by group at each stage of a funnel, not just at the offer.
Four-fifths rule
A rule of thumb from the US Uniform Guidelines on Employee Selection Procedures. Take the group selected at the highest rate. If any other group is selected at less than 80 percent of that rate, it is treated as evidence of adverse impact worth investigating. It is a screening heuristic rather than a legal verdict, and small samples make it noisy. It remains the number most bias audits report against.
Automated Employment Decision Tool (AEDT)
The term New York City Local Law 144 uses for a computational process that substantially assists or replaces discretionary decision-making in hiring or promotion. If a tool meets the definition, the employer owes an annual independent bias audit, a published summary of results, and advance notice to candidates. The definition turns on how much the tool decides, which is why human-in-the-loop design matters legally and not only ethically.
Bias audit
An independent review of a selection tool's outcomes across protected categories, published as a summary of impact ratios and selection rates. Under NYC Local Law 144 it must be conducted by an independent auditor and repeated annually before the tool is used. An audit is about outcomes on your data, so it is a recurring obligation rather than a certificate a vendor can hand you once.
Candidate notice
The disclosure a candidate is owed before an automated tool assesses them, covering that it is being used, what it evaluates and often how to request an alternative. Requirements differ by jurisdiction: NYC sets a notice period, Illinois has its own rules for AI-analysed video interviews, and the EU AI Act adds transparency duties for high-risk uses. Notice is cheap to get right and expensive to skip.
Bias mitigation
The set of practices used to keep irrelevant characteristics out of an evaluation. Redact identifying details before analysis, score against fixed criteria, keep questions job-relevant, and monitor outcomes afterwards. No single technique makes a process fair, and claims of a bias-free tool should be read with suspicion. What can be done is to remove obvious leakage and then measure what happens.
Diversity, equity and inclusion (DEI)
The organisational practice of widening who gets considered and making sure the process treats them consistently once they are. In selection terms it lives or dies on the mechanics: where you source, what you ask, and whether every candidate is scored the same way. Structured rounds and identity redaction are the parts a hiring tool can actually contribute.
PII masking
Redacting personally identifying details from a resume or profile so a reviewer sees the substance rather than the name, address or other identity markers. Expert Hire applies server-side resume masking before reports leave the pipeline, and the Chrome extension masks PII on resumes before they are shared. A masking action counts the same as a resume scan, 1 Hire Credit.
EU AI Act
The European Union's risk-tiered regulation of AI systems, which classifies employment and worker-management uses as high risk. High-risk classification brings obligations around risk management, data governance, human oversight, logging and transparency. If you assess candidates inside the EU, the relevant question is not whether the rules apply but which obligations land on you as a deployer.
Integrity and proctoring
Remote assessment created a cheating problem and a credibility problem at the same time. These are the terms for how it gets handled without turning a candidate into a suspect.
Proctoring
Monitoring a candidate during an assessment to establish that the work is theirs. Expert Hire offers browser-based proctoring for lower-stakes rounds and a cross-platform desktop application for strict rounds, which brings video, the code editor and security monitoring into one place. Proctoring adds no credits to an assessment and unlocks at Growth and Scale.
Malpractice detection
Detecting known AI-assistant overlay tools, signs of impersonation and tab-switching during a live round, plus post-interview video analysis. Every signal is surfaced to a human reviewer next to the transcript and the code, and a flag is never an automatic verdict. The reportcard carries a malpractice status so a clean round is stated explicitly rather than assumed.
Impersonation
Someone other than the applicant sitting the assessment, whether a paid proxy or a colleague on a second screen. It is the failure mode that quietly invalidates a whole remote funnel, because the pipeline still looks healthy. Expert Hire surfaces signs of impersonation to a human reviewer rather than acting on them automatically.
Tab-switch signal
A count of how often a candidate left the assessment window during a round, shown in the proctoring banner and carried onto the report. It is context rather than a verdict: a handful of switches during a long coding problem reads differently from a burst right before a correct answer appears. That judgment stays with the reviewer.
Desktop proctoring app
A cross-platform desktop application used for strict remote proctoring, bundling video, the code editor and security monitoring in a single controlled environment. It exists because browser-level monitoring can only see so much, and high-stakes rounds sometimes need more. It is distributed alongside the web app and the Chrome extension.
Platform, pricing and operations
Terms that come up in a procurement call or an implementation thread: how the workspace is structured, how usage is charged and what connects to what.
Applicant tracking system (ATS)
The system of record for candidates and requisitions, where applications are collected, sorted and moved through stages. Expert Hire ships a built-in kanban pipeline with Pre-contact, Contacted, Screened and Shortlisted stages, source tags, search and export, usable as your own ATS on every plan. If you already have one, Greenhouse is available today in early access with bidirectional sync, and Bullhorn, Lever, Workday and Vincere are on the roadmap.
Hire Credits
The single monthly pool a hiring workspace draws from, so you are never billed per seat and team members stay unlimited on every tier. Per-action cost is fixed. A resume screen is 1, a human interview with copilot 2, a notetaker bot 2, a prompt assessment 3, a coding test 3 and an AI interview 5. Candidate matching, proctoring and reports are 0. Unused plan credits roll over one month, and the 30-day free trial includes 75 credits with nothing charged for 30 days.
Prep Credits
The job-seeker equivalent of Hire Credits, bought once as a pack with no subscription attached. A resume review or Psychometric Test costs 1, a coding or prompt-engineering assessment 3, and a mock interview, VC Pitch or full question-bank unlock 5, and every purchase comes with a downloadable GST invoice. It exists so a candidate can prepare without signing up to a recurring plan.
Client workspace (sub-company)
A fully separated client workspace inside a staffing agency's single account, with its own roles, pipeline, skills and branding. The product calls these Clients. The API and your bill call them sub-companies. Data never pools across clients, which is what makes one login safe to use across competing accounts. Client workspaces are included on every plan.
White-label
Running the candidate-facing product under your own brand rather than ours. That covers your custom domain with automated TLS, your logo, brand name and favicon. Candidate email is sent from a DKIM-signed domain of yours, so it arrives as you. It can be delivered standalone or embedded in your own product with per-partner theming. Basic white-label unlocks at Growth and the full version at Scale.
Notetaker bot
A bot that joins a scheduled external call, with participants notified of recording, then records, transcribes and scores the conversation against the job. It costs 2 credits per external call joined and is available on Starter and up. It turns calls that were happening anyway into scored, searchable evidence.
Instant meeting room
A persistent, on-demand room for conversations that happen without a calendar invite, launched straight from the recruiter dashboard. The session is recorded and auto-classified into a titled report, and a live copilot suggests follow-up questions during the call. Meetings are a first-class item in the workspace sidebar rather than a side feature.
Model Context Protocol (MCP)
An open protocol that lets an external AI assistant call a product's tools directly. Expert Hire exposes dozens of recruiting tools over a secure OAuth-protected remote MCP server, so hiring can be run from Claude, with ChatGPT support in beta. Financial actions, team-admin changes and destructive operations are deliberately excluded from that surface.
Mia
The recruiter-side copilot inside the Expert Hire workspace, reached through Chat with Mia in the sidebar. Mia's tools are the same ones exposed over MCP, so the assistant in the product and the assistant in your chat client are working from one surface. Mia is the recruiter's copilot; the AI who interviews candidates is Ethan.
Ethan
The AI interviewer candidates actually meet in an AI round. Ethan runs the round in voice, asks follow-ups based on the candidate's answers, and grades communication, confidence and technical depth. Naming the interviewer is deliberate: a candidate should know exactly what they are talking to.
Psychometric Test
A psychometric assessment of about 7 minutes and 40 questions that returns a work-style profile. It describes personality at work in plain language and the roles someone would naturally thrive in. It also says whether they fit startup or enterprise, product or service. Past profiles are stored with a label and a best-fit line. It is a self-knowledge tool for candidates, not a screening filter.
Candidate experience
What the process feels like from the applicant's side: how long they wait, how clearly they are told what happens next, and whether they get anything back. It is a recruiting metric because rejected candidates talk, and because the best applicants have other offers moving faster. Rounds that run 24/7 on the candidate's own schedule and language exist mostly for this reason.
AI foundations
The underlying machine-learning vocabulary, defined the way it actually matters in a hiring context rather than in the abstract.
Large language model (LLM)
A model trained on very large text corpora that predicts and generates language, and the component behind conversational AI interviewers, summaries and transcript analysis. Its strength is fluency across open-ended input; its weakness is that fluency is not the same as correctness. In hiring that gap is closed by grounding output in a transcript and a fixed rubric.
Natural language processing (NLP)
The broader field concerned with getting machines to work with human language: transcription, entity extraction, classification and summarisation. In a hiring pipeline it is what turns spoken answers into text that can be searched, filtered and scored. Most of the boring, high-value work in the stack is NLP rather than generation.
Generative AI
AI that produces new content rather than only classifying existing content, including text, code, audio and images. In recruiting it drafts job descriptions, generates round questions and produces written summaries. Everything it generates still needs a human owner before it reaches a candidate.
Retrieval-augmented generation (RAG)
A pattern where a model retrieves relevant material from a trusted source before generating an answer, instead of relying only on what it absorbed in training. It keeps output anchored to real documents, which matters when the answer needs to reflect a specific job, rubric or transcript rather than a plausible average. It is the standard mitigation for hallucination.
Hallucination
When a model states something confidently that is not true. In hiring the risk is specific and serious: a summary that attributes a skill nobody demonstrated becomes a decision nobody can defend. The countermeasure is grounding, which is why every score should point back to the exact evidence it came from.
Computer vision
Machine interpretation of images and video. In assessment it appears in proctoring and post-interview video analysis, where it is one input among several. It should never be the sole basis for an integrity call, because appearance, lighting and disability all affect what a camera sees.
Tokenization
Splitting text into the smaller units a model actually processes. It is the reason model limits and costs are expressed in tokens rather than words, and why a long transcript behaves differently from a short one. Mostly invisible until you are budgeting for volume.
Zero-shot learning
A model handling a task it was never given specific training examples for. It is why a new or unusual role can be assessed without first assembling a labelled dataset for it. The trade-off is that zero-shot performance is harder to validate, so it belongs behind a rubric and a human reviewer.
Artificial general intelligence (AGI)
A hypothetical system that could learn and apply knowledge across arbitrary tasks at human level. It is not what any recruiting product ships today, and the term is worth keeping distinct from the narrow, task-specific models that actually run in production. Treat it as a research horizon rather than a product category.
Round mode, rubric, reportcard, readiness score.
Four of the terms on this page are one chain. Read them in order and you know how an evaluation holds up. Here is the artifact at the end of it.
What you are looking at: a reportcard for one candidate after Round 2. An overall score of 84 with a Strong fit for role chip, a skill breakdown scored from the transcript and the code, the full 16:41 recording, a speaker-separated transcript of 96 lines you can search and filter, and a malpractice line reading None with no tab switches flagged. Tabs across the top for Overview, Analysis, Submissions and Resume, with Share and Download PDF. All names and figures shown are sample data.
Round mode
You pick one of five: Resume Screening, AI Interview, Human-Led Interview, AI Prompt Assessment or Coding Test. Inside an AI interview you then pick one of eight round types, from HR Round to System Design. The choice sets what evidence the stage produces.
Rubric
You pick the skills for the job and set a weightage on each one, and you set up the JD. Answers are scored against both, and the questions adapt to what the candidate says. That is what makes a coding test comparable to an interview.
Reportcard
One report per candidate: an overall score, a verdict line, a skill breakdown, the recording, a searchable transcript and a malpractice status. Reports are never metered.
Readiness score
Every round rolls into one comparable number per candidate, per student, or aggregated into a batch readiness score for a campus. That is the number a shortlist is defended with.
A score with no evidence attached is an opinion with a number on it. The methodology page covers what each round measures and how a score traces back to the transcript, the code, your skills and your JD.
Read the methodologySome of these terms carry an obligation.
Four jurisdictions have written automated hiring into statute, and each one introduced vocabulary the rest of the industry now borrows. These guides cover what applies, to whom, and what you have to produce. They are not legal advice.
New York City, Local Law 144
The law that made Automated Employment Decision Tool a term of art. It requires an independent annual bias audit, a published summary and advance notice to candidates before an in-scope tool is used.
Illinois, AI Video Interview Act
Rules specific to AI analysis of video interviews: explain how the AI works, get consent before it is used, limit who sees the recording and delete it on request.
California, employment fairness
Automated decision systems in employment are treated as an extension of existing discrimination law, with expectations around outcome testing and retention of records.
European Union, AI Act
Employment and worker management sit in the high-risk tier. That brings duties around risk management, data governance, logging, human oversight, and telling people what they are being assessed by.
Not sure which of these reaches you? The US state-by-state guide tracks what is in force, what is pending and what it asks of an employer.
State-by-state guideWhere each of these gets explained properly.
A definition is a starting point. These pages carry the detail behind the terms that matter when you are choosing how to screen people.
Scoring methodology
Readiness score, rubric and reportcard in full: what each round measures and how a number traces back to evidence.
ReadMalpractice detection
Proctoring, overlay-tool detection and impersonation signals, and why each one goes to a human reviewer instead of an automatic verdict.
ReadCoding tests
What a real work sample looks like: sandboxed execution, hidden test cases, per-case runtimes and a complexity read.
ReadNYC Local Law 144
The AEDT definition, the annual bias audit, the notice period, and a questionnaire for working out whether you are in scope.
ReadAI interview platform
Where the vocabulary becomes a product: round modes, AI and human interviews, and the report that comes out the other end.
ReadQuestion bank
Fifteen role guides with the prompt, the signal a strong answer carries, the follow-up probes and the red flags.
ReadComparisons
How Expert Hire lines up against the tools you are probably also looking at, feature by feature.
ReadBlog
Longer writing on hiring practice, assessment design and what is changing in the regulation around both.
ReadThe questions behind the definitions.
Where the vocabulary usually causes an argument.
Still have a question? Book a demo and we will walk through it with your process in front of us.
Knowing the words is the easy half.
Run a structured round and see what the vocabulary actually produces: one rubric across every stage, a reportcard with the transcript attached, and a readiness score you can defend in a debrief.
30-day free trial with 75 Hire Credits. Card required, nothing charged for 30 days.