AI Recruiting Terms: A Plain-Language Glossary for Recruiters
What the AI words in vendor demos, ATS release notes and compliance emails mean for your recruiting work

AI recruiting terms now turn up in vendor demos, ATS (applicant tracking system) release notes, compliance emails, and hiring manager questions, often unexplained. Some describe how a tool searches or ranks people. Others carry legal duties, and a few are marketing labels with no fixed meaning.
This AI glossary groups the terms by where you'll meet them. Each entry gives a plain definition, then what the term changes in sourcing, screening, or compliance work.
Key takeaways
- Same word, different meanings: Vendors define terms such as matching score and AI agent in their own ways, so watch how a feature behaves on your roles.
- Rules versus models: Knockout questions and workflow automation follow fixed rules, while matching and drafting features rely on trained models.
- Some terms carry legal weight: Automated employment decision tool, bias audit and automated-decision system have specific meanings in certain U.S. cities and states.
- Questions beat jargon: The vendor-demo table near the end turns key terms into questions to ask before you buy.
Core AI terms recruiters hear
These broad terms appear on nearly every product page, and the rest of the AI terminology in vendor materials builds on them.
Artificial intelligence (AI)
Artificial intelligence (AI) is software that performs tasks that would otherwise require human judgment, such as reading text or making recommendations. SHRM's 2026 research found that 39% of the organizations it studied had implemented AI in HR, with recruiting leading at 27%. Since the label covers so much of AI recruiting, ask which task a product performs and whether its output is a suggestion or a decision.
Machine learning and training data
Machine learning is a type of AI that learns patterns from past examples instead of following rules a person wrote. Training data is the set of examples it learned from, such as resumes, job descriptions, and past hiring outcomes. A model trained on your past hires can help AI hiring tools predict fit, but it also learns your old habits. One trained on a single industry may misread candidates from another.
Generative AI and large language model (LLM)
Generative AI creates new content, such as text or images, in response to an instruction. A large language model (LLM) is generative AI trained on huge amounts of text. It powers the drafting and summarizing features in recruiting software. LLMs predict likely wording rather than look facts up, so treat their output as a first draft that someone reads before it reaches a candidate.
Natural language processing (NLP)
Natural language processing (NLP) is the branch of AI that lets software interpret human language. It sits behind resume parsing and search bars that accept full sentences. It can still misread abbreviations and company-specific job titles.
Prompt
A prompt is the instruction you type into an AI tool to get a result. Naming the role, location, must-have skills, and tone gives the tool more to work with than a one-line request. A prompt that works for one role can be reused for the next.
Hallucination
A hallucination is confident-sounding AI output that's false or made up. NIST's Generative AI Profile, a July 2024 cross-sector risk framework, calls this confabulation: confidently stated but false content that can mislead people. In recruiting, it might be an invented certification in a candidate summary or a salary figure nobody gave.
Search and matching terms
These terms appear when a tool promises to find or rank candidates for you.
Semantic and natural-language search
Semantic search finds results by meaning rather than exact word match. Natural-language search lets you type a request as an ordinary sentence, such as "bilingual customer success managers in Texas," and the tool builds the filters. Together they surface people who describe the same work in other words. Check the filters, because a tool can drop a requirement or treat a preference as a must-have.
Boolean search
Boolean search combines keywords with operators such as AND, OR, and NOT to include or exclude exact terms. It offers precise control but can miss people who describe a skill differently. Running Boolean search strings alongside semantic search lets you tighten the pool with one and widen it with the other.
Matching score and ranking
A matching score is a number or label showing how closely a candidate fits a job, and ranking is the order those scores produce. Each vendor weighs skills, titles, and experience differently, so two tools can score one person differently. With a large pool, the ranking decides who gets reviewed, so sample a few lower-ranked candidates when the criteria are new.
Explainability
Explainability is a tool's ability to show why it produced a result in terms a person can follow. In matching, it appears as short why-match reasons next to each candidate, such as experience managing ICU staff. Those reasons let you correct a bad match and explain a shortlist to a hiring manager or client.
Screening terms
These terms show up in ATS settings, application forms, and high-volume hiring pitches. They mark the points where software filters people out.
Resume parsing
Resume parsing is software that pulls contact details, job titles, dates, and skills from a resume into structured fields. Resume parsing tools can mangle tables, columns, and unusual headings. A record missing a job title can then fail a filter it should have passed.
Knockout questions
Knockout questions are fixed-answer questions that remove applicants who don't meet a basic requirement, such as a required license. They're rule-based rather than AI, but they often sit inside AI screening flows. A question about a nice-to-have, or one qualified people misread, can remove candidates nobody meant to lose.
Automated screening and auto-reject
Automated screening uses software to sort, score, or filter applications before a person reads each one. Automated candidate screening can combine parsing, knockout questions, and AI scoring, so record which rules and models run at each stage. Auto-reject rejects applicants who fail a rule without human review. A parsing error can then turn away a qualified person unnoticed, which is why it suits only clear-cut requirements.
AI interview
An AI interview is a screening step where software asks questions through one-way video or a text chatbot. Some tools only record and transcribe, while others score the answers. Scoring raises fairness and notice questions, and candidates who can't use the format need an alternative.
Human in the loop
Human in the loop means a person reviews an AI tool's output before it affects a candidate. It only works if that person has the time, information, and authority to disagree. In a 2025 University of Washington study using simulated AI, 528 participants mirrored its recommendations and followed severely biased picks around 90% of the time. It was a lab study, not recruiters at work, but it shows why reviewers need to see the reasons behind a pick.
Assistant, agent and automation terms
These terms describe AI that does work for you, not just searching or scoring.
AI copilot, AI agent and agentic AI
An AI copilot or assistant helps with a task when you ask, such as drafting a message, and leaves the next step to you. An AI agent plans and carries out a series of steps toward a goal with limited human input. Agentic AI is the broader label for systems built this way. In RedThread Research's study of people analytics vendors, as reported by SHRM, 70% prioritize AI copilots, and 49% invest mainly in agents.
The same SHRM report found 41% of those vendors have little or no meaningful agentic AI capability. Those figures describe people analytics vendors, not every recruiting tool. Before switching on an agent, check what agentic AI tools do on their own and where they stop and need approval.
Workflow automation
Workflow automation carries out set steps when a trigger happens, such as moving a candidate forward after a form is submitted. It follows rules you define and needs no AI. Knowing which steps are rules and which are models helps when a client asks how a decision was made.
Recruiting chatbot
A recruiting chatbot answers candidate questions, collects details, or schedules interviews through text or chat. Older ones follow scripts, while newer ones use an LLM to write replies. A recruiting chatbot can cover after-hours questions, as long as candidates can reach a person and someone reviews its pay-related answers.
Integration and application programming interface (API)
An integration lets two systems share data, such as a sourcing tool and your ATS. An application programming interface (API) is the set of rules software uses to request and send that data. A one-way sync can leave your ATS out of date.
MCP
Model Context Protocol (MCP) is an open standard for connecting AI assistants to other tools and data. It could let an assistant read your ATS or calendar without a custom integration for each. Because each connection opens real records, understanding how MCP works helps you decide what an assistant can read or change.
Fairness, compliance and data terms
Several of these AI recruiting terms carry legal meaning. Rules depend on where each role is located, so check with counsel before relying on any definition here.
Algorithmic bias and adverse impact
Algorithmic bias is a pattern in which an AI system's outputs consistently favor or disadvantage certain groups. In a 2024 University of Washington study, three open-source models favored white-associated names 85% of the time and Black-associated names 9% of the time. Those were research models, not commercial hiring tools.
Adverse impact is a difference in selection rates that disadvantages a protected group, even when nobody intended it. You spot it by comparing how often groups pass a step such as a screen, and counsel should confirm how it's measured.
Proxy
A proxy is a data point that stands in for a protected characteristic a tool isn't supposed to use. Illinois Public Act 103-0804, effective January 1, 2026, bars Illinois employers from using zip codes as a proxy for protected classes. A model that never sees age can still infer it from graduation year, so know every field a tool reads.
Automated employment decision tool (AEDT) and bias audit
New York City calls automated employment decision tools (AEDTs) the term for automated tools that employers and employment agencies use in employment decisions. A bias audit checks whether a tool's results differ across demographic groups. Under NYC Local Law 144, an AEDT can be used only if it has had a bias audit within one year and the audit information is public.
The same law requires notices to candidates or employees, and the city began enforcing it on July 5, 2023. Whether a given tool is covered is a question for counsel.
Automated-decision system
California calls an automated-decision system software that makes or helps make employment decisions. Under California's civil rights rules, using one may violate California law if it harms applicants or employees based on protected characteristics, effective October 1, 2025. The same rules require employers to keep employment records, including automated-decision data, for at least four years. If a vendor holds that data, its retention settings need to match.
Candidate notice
A candidate notice tells applicants or employees that AI or an automated tool is part of the process. For example, Illinois's amended law requires employers to notify employees when they use AI. Content and timing differ by location, so keep a record of what each candidate received.
Data privacy and opt-in data
Data privacy is handling personal information so it's used and shared only as permitted. NIST's profile lists data privacy as a generative AI risk, including leakage of personal information. Before resumes go into an AI tool, find out whether the vendor trains models on them.
Opt-in data is information a person knowingly agreed to share for a stated purpose, such as a profile created to hear from recruiters. Unlike scraped or purchased data, it gives you a clear answer when a candidate asks how you found them.
Ask about these AI recruiting terms before your next vendor demo
Vendors use AI recruiting terms loosely, so the same questions help you compare demos. Ask each vendor to run a search for one of your open roles, and note which answers come with evidence.
Terms to ask about in a vendor demo
| Term | Question to ask |
|---|---|
| Matching score | Which inputs raise or lower a candidate's score? |
| Explainability | Can I see why each candidate was matched? |
| Training data | What data trained the model and how current is it? |
| Human in the loop | Where must a person approve before a candidate is affected? |
| Auto-reject | Can I turn off automatic rejections at any stage? |
| Bias audit | Has the tool had a bias audit and can I see the summary? |
| Integrations and APIs | Does data sync both ways with our ATS? |
See how Rotto explains every match
Book a demoFrequently asked questions
Do recruiters need to understand how AI models work?
You don't need the math behind a model to use AI tools well. You do need to know what data a tool uses, what it produces and where a person reviews the result.
How can a recruiting team keep its AI vocabulary consistent?
Agree on a short shared list of the AI recruiting terms your tools use and what each means in your process. Note the vendor's definition beside yours when they differ, and share the list with hiring managers or clients.
How often do AI recruiting terms change?
New AI recruiting terms appear as vendors release features and lawmakers write new rules. Legal definitions change only through new laws and formal rulemaking, so they move more slowly than vendor labels.
About the author

Founder & CEO of Rotto, building tools that help tech recruiters source better candidates, faster.





