Agentic AI Recruiters: What to Do When an AI Agent Calls You for a Pre-Screen

Hirelytica Team • • 12 min read

TL;DR

63% of job seekers in 2026 have already been interviewed by an AI agent, up 13 points in six months. 70% were not told it was AI. 38% have withdrawn from a hiring process because of it. By mid-2026, analysts expect roughly 80% of high-volume hiring to start with an AI-led voice screen.

The agent is not judging your voice. It is transcribing your answers and scoring the transcript against a rubric. The candidates who win these screens give specific PAR (Problem, Action, Result) answers with named tools and real metrics, treat the agent like a structured phone screen, and never — ever — try to use a second AI to feed them live answers.

Your phone rings. The voice on the other end is calm, friendly, asks a thoughtful question, and waits for your answer. It sounds like a recruiter doing a phone screen. It is not. It is an agent — software that will conduct the entire 15-minute conversation, transcribe every word, score you against a rubric, and either advance or quietly reject you, all before a human being ever sees your name. In 2026 this is now the modal first round for high-volume roles, and most candidates walk into it blind.

The Agentic AI Reality in 2026

Agentic AI in recruiting has moved from pilot to default in under 18 months. Here is the data that explains why every job seeker now needs a playbook for these screens:

63% of job seekers have now been interviewed by an AI, up 13 percentage points in six months (Greenhouse 2026 Candidate AI Interview Report)
70% of those candidates were not told AI was involved in their interview
38% of candidates have already withdrawn from a hiring process because it included an AI interview; another 12% would walk if forced
46% of organisations are using or planning agentic AI in talent acquisition; 52% of TA leaders plan to deploy autonomous agents by end of 2026
~80% of high-volume hiring is expected to start with an AI-led voice screen by mid-2026
AI usage in HR jumped from 26% to 43% of teams in a single year (2024 to 2025)

Sources: Greenhouse 2026 Candidate AI Interview Report (2,950 active job seekers), Intervuebox 2026 industry survey, Aisera 2026 Recruiting Guide, SHRM Talent Trends 2025.

Two things are happening at once. AI agents are becoming the default first contact for thousands of roles. And candidates are arriving at those calls with no preparation, no understanding of how the system actually scores them, and no idea what their rights are. The result is a mass mismatch: motivated candidates losing roles to better-prepped peers who simply read up on the format.

What an Agentic Recruiter Actually Is

An agentic AI recruiter is software that takes autonomous action across recruiting workflows: scheduling, screening, interviewing, scoring, and shortlisting — not just answering questions. The shift from chatbot to agent is the shift from “give me information” to “run this part of the process for me.”

The platforms you will most likely encounter

Paradox (Olivia): The frontline-volume leader. Used by FedEx, Unilever, McDonald's. Handles 100+ simultaneous candidate conversations and compresses 5-7 day workflows into under 48 hours.
HireVue: The Fortune 500 incumbent. Structured video and voice interviews, used at enterprise scale across multiple departments and locations.
Mercor: AI-native talent layer. $10B valuation as of October 2025. Heavily used for tech and contract talent.
Sapia.ai: Text-based behavioural interviews focused on removing visual bias. Common at organisations with 200+ employees prioritising DEI.
HeyMilo / Carv / Phenom: Voice-first agents that run the call directly, in 50+ languages, around the clock.
Ezra (now Greenhouse): Greenhouse acquired Ezra AI Labs in 2026 to bring conversational AI directly into its ATS. Expect this experience inside any Greenhouse-powered process.

The conversational AI market hit roughly $41 billion in 2026, with the HR and recruiting slice growing at a 25% CAGR — the fastest of any vertical. Translation: the procurement budget for these tools is now enormous, and almost every ATS will ship native agent capability within 12 months.

How an AI Pre-Screen Actually Works

The format is consistent across most platforms. Knowing the structure removes 80% of the surprise.

The five-stage AI pre-screen

1. Invitation: You receive a link via email, SMS, or directly in the ATS. No scheduling step — you complete it whenever you want, often within 48 hours of applying.
2. Identity check: Some platforms now require a government ID upload and liveness selfie before the call. Greenhouse Real Talent (in partnership with CLEAR) is the most visible example.
3. The interview: 10-20 minutes. The agent asks 5-8 questions, listens to your spoken answers, and generates follow-ups based on what you say. Modern systems handle natural pauses, accents, and code-switching.
4. Transcription + scoring: Your answers are transcribed and scored against a structured rubric the employer set: competencies, keywords, completeness, sentiment markers.
5. Outcome routing: A summary plus your transcript lands in the recruiter's inbox with a recommended decision. Sometimes the agent advances or rejects you autonomously.

One critical thing to internalise: the agent is not judging your voice tone, your microexpressions, or how confident you sounded. The transcript is the artefact. By the end of 2026, the recorded transcript will have replaced video as the primary record of what happened. That is good news for candidates who interview better in writing than in performance — if you have substance, the format actually plays to your strengths.

What the Agent Is Actually Scoring

This is the part most candidates get wrong. They prepare for the AI screen as if it were a chat with a human recruiter, focusing on rapport and warmth. The agent does not score those things. It scores against a rubric.

The four signals every agent looks for

1. Competency hits: Did your answer contain evidence of the named competency (e.g., “stakeholder management,” “debugging in production,” “running a P&L”)? One example beats five adjectives.
2. Keyword overlap with the JD: Specific tools, methods, frameworks, or technologies named in the job description that you also named in your answer. Mirror their vocabulary.
3. Quantified outcomes: Real numbers with context. “Cut churn from 18% to 11% on 2,400 SMB accounts over four months” outscores “significantly improved retention.”
4. Answer completeness: Did you actually answer the question, or did you talk around it? Most agents flag “non-responsive” answers automatically.

Notice what is not on this list: vocal warmth, charm, or whether the agent “liked” you. Those things matter in a human screen. They do almost nothing in an agentic one. If you have ever bombed a phone screen because the recruiter was distracted or grumpy, the AI version is genuinely fairer in some structural ways. It does not get bored, it does not skim your answers, and it does not unconsciously prefer candidates who sound like the last good hire.

For more on the underlying CV-side signals these systems pick up, see our piece on how recruiters spot AI-written CVs — the same specificity tests apply to voice answers.

The Pre-Screen Playbook: How to Prep in 90 Minutes

You do not need to overhaul your interview prep. You need to recalibrate it for a structured-rubric format. Here is the minimum-viable prep you should do before any AI pre-screen.

Step 1: Build 4-6 PAR stories (45 minutes)

Pick 4-6 of your strongest career moments and write them out in PAR format: Problem (the specific situation), Action (what you did, with named tools), Result (a measurable outcome with context). Each story should fit comfortably into a 90-second spoken answer.

Make sure your stories cover the core competencies in the JD. If the role mentions stakeholder management, ownership, customer empathy, and analytical rigour, you need at least one story per competency.

Step 2: Mine the JD for keywords (15 minutes)

Pull out the specific tools, frameworks, methodologies, and outcomes named in the job description. These are the keywords the agent's rubric will be looking for. Work them naturally into the language of your answers — not in a stuffed way, but the way someone who actually uses them daily would.

For deeper detail on this approach, see the ATS optimisation guide. The same keyword logic that gets your CV through ATS screening applies to your spoken answers in front of an agent.

Step 3: Test your setup (15 minutes)

Quiet room. Wired headphones if possible (Bluetooth introduces lag and dropouts that can make the agent mishear you). Wired internet if your Wi-Fi is unreliable. Phone on silent and somewhere you cannot accidentally pick it up.

Do a 30-second test recording on your phone first. Listen back. If your voice is muffled, garbled, or echoing, fix it before the call. The agent will transcribe what it hears — if it mis-transcribes, you score badly through no fault of your answers.

Step 4: Rehearse out loud, twice (15 minutes)

Read each PAR story out loud once. Then say it again from memory in your own conversational voice. Do not memorise word-for-word — the agent will detect a recited script. The goal is to be fluent enough that you do not stumble, but natural enough to sound like a person.

If the platform offers a practice question (HeyMilo, Sapia, and Ezra typically do), use it. The first 30 seconds of any AI call is when most candidates underperform because the format is unfamiliar.

During the Call: The Five Rules

When the agent calls, the things that win are different from a human screen. Here are the five rules that consistently move you up the rubric.

The five in-call rules

1. Take a beat before answering. A 2-3 second pause to think reads as composure, not hesitation. Agents do not penalise pauses; they penalise rambling.
2. Lead with the specific. Open every answer with the company, the role, the year, the tool, or the team size. “In Q3 2024 at Acme, I led a team of four PMs through…” That single sentence ticks half the rubric.
3. End with the number. Always close a PAR answer with the result and its context. “…which moved monthly active users from 12k to 19k over six weeks.”
4. Ask for repeats if needed. Modern agents handle “Sorry, can you repeat the question?” smoothly. Better that than answer the wrong question.
5. Stop when you have answered. Do not pad. Padding dilutes your competency hits and increases the risk of contradicting yourself in the transcript. 60-90 seconds per answer is the sweet spot.

The Trap: Using AI to Cheat the AI

You will be tempted to run ChatGPT in another window during the call and feed yourself answers. Do not. The detection technology has caught up faster than the cheating tools, and the consequences are now severe.

What detection systems now catch

Unnatural pause patterns: Real humans pause mid-thought; AI-fed candidates pause uniformly to read the screen.
Filler-word frequency mismatch: Suspiciously few “ums” and “you knows” for a spontaneous answer.
Vocabulary signatures: The same AI-flavoured words (“leveraged,” “robust,” “intricate”) that get CVs flagged also flag spoken answers.
Liveness and audio-authenticity checks: Some platforms flag re-routed audio, voice clones, and synthetic speech.
Multi-window detection: Some platforms flag tab-switching and unusual mouse/keyboard activity during the call.

The numbers behind the crackdown: 59% of hiring managers now suspect candidates of using AI to misrepresent themselves; 1 in 3 has caught a candidate using a fake identity or proxy in an interview; 17% have personally encountered deepfake technology in a video interview. When InCruiter rolled out its deepfake detection in early 2026, it found fraudulent activity in 25-30% of flagged sessions — nearly double what experienced human interviewers had previously caught.

For the wider context on why this matters and how identity verification is rolling out across major ATS platforms, read our piece on the fake candidate crisis and identity verification.

Real Talk: Your Rights and the Disclosure Problem

There is a part of this conversation that the AI vendors do not want to lead with. The Greenhouse 2026 Candidate Report found that 70% of candidates were not told AI was involved in their interview. That is the disclosure crisis, and it is not okay.

What candidates actually want

From the same Greenhouse survey of 2,950 active job seekers, only 19% want less AI in the hiring process. The majority want the same or more. But with guardrails:

44% want upfront disclosure that AI is involved
39% want a clear explanation of what the AI is measuring
46% want the option to request a human interview instead
67% are comfortable with AI doing initial screening — if a human makes the final hiring decision (SHRM 2025)

Here is what you can actually do. Before the call, email the recruiter and ask three questions: Will this interview involve AI? What is the AI assessing? Is there a human-interview alternative if I prefer? Reasonable companies will answer all three. Companies that refuse, dodge, or say “our process is confidential” are telling you something useful about their culture.

The EU AI Act's August 2026 deadline mandates human oversight for high-risk employment AI systems. Even outside the EU, “human-in-the-loop” for consequential hiring decisions is becoming the baseline expectation. If a company tells you no human will look at your interview before a decision is made, that is now an outlier — and a red flag.

The Outcomes Data: What Actually Happens After

The Greenhouse report tracked what happened to candidates after they completed an AI interview. The numbers are sobering, and useful.

After the AI interview

28% moved forward to the next round
13% received a formal rejection
51% never heard back
53% of candidates who advanced from an AI screen passed the next human round, vs 29% from resume-only review
Completion rates for AI phone screens hit 70%+, vs 42% video interview dropout

Sources: Greenhouse 2026 Candidate AI Interview Report; HeyMilo and Carv platform data, 2026.

That 51% “never heard back” figure is a real problem and not something you can fix from the candidate side. What you can fix is the input. Candidates who do well in AI screens have a meaningfully higher pass rate in the human rounds that follow — almost 2x — because the rubric-based format rewards substance, and substance compounds. If your AI screen lands well, the rest of the process becomes easier.

When You Should Walk Away

Most AI screens are worth completing. Some are not. Use these signals to decide.

Walk away if…

The company refuses to confirm whether AI is involved or what it is measuring.
The platform demands invasive permissions (full screen recording, browser history, microphone always-on) you would not give a human.
The role is junior, the screen is 45+ minutes, and the agent is testing skills you would only need on the job.
No human will see your transcript before a hiring decision is made (rare in 2026, but happens at the bottom of the high-volume market).
The role has obvious ghost-listing signals — reposted for months, no recruiter named, no clear hiring manager. See our ghost jobs analysis.

If none of those apply, do the screen. The maths is firmly in your favour: you can complete an AI pre-screen at 11pm in your kitchen, in 15 minutes, with no scheduling overhead. That is the cheapest route into a hiring funnel that has otherwise become brutal — see why LinkedIn Easy Apply is broken for what the alternatives look like.

Where Hirelytica Fits

The hardest part of preparing for an AI pre-screen is having the raw material at your fingertips: the right PAR stories, the right metrics, the right tools, surfaced quickly for each role you apply to.

Structured CV Library: Every project, achievement, metric, and tool from your career stored as queryable data. The PAR stories you need for any AI screen are already indexed.
Per-role surfacing: Pull the most relevant achievements for each specific role, with the keywords that match the JD already aligned.
For multi-shaped careers: If your career spans IC and management, hardware and SaaS, in-house and consulting, you need different stories for different roles. Hirelytica makes that switch instant.
Human-centred with AI assistance: AI helps you prepare; humans make the decisions that matter. That is the philosophy — on both sides of the interview.

For more on why senior, multi-shaped careers struggle most with the broken funnel that AI screens are stitching together, see our piece on why recruitment is broken (the data).

Frequently Asked Questions

What is an agentic AI recruiter and how is it different from a chatbot?

An agentic AI recruiter is an autonomous system that runs full segments of the hiring process on its own: scheduling, screening, conducting voice or chat interviews, scoring answers, and pushing shortlists to a human recruiter. A chatbot answers questions; an agent takes actions, makes judgements, and progresses or rejects you with limited human oversight. Paradox's Olivia, Mercor, Sapia, HeyMilo, and Ezra (acquired by Greenhouse in 2026) are the platforms candidates encounter most often.

Should I refuse an AI pre-screen interview?

It is your call, but the data says refusing usually costs you the role. 63% of job seekers have already done one, 38% have withdrawn from a process because of one, and 12% would withdraw if forced. If the role is one you genuinely want, do the AI screen and use the option to request a human interview later. If the company refuses to disclose what AI is measuring or to offer any human alternative, that is a culture signal worth taking seriously.

How are AI agents actually scoring my pre-screen answers?

Modern voice and chat agents transcribe what you say, then score the transcript against a structured rubric the employer set. They look for specific competencies (e.g., “gives examples with measurable outcomes”), keywords matched to the JD, sentiment markers, and answer completeness. They are not assessing your voice tone or face in any meaningful way; the transcript is the artefact. That is why specificity, real metrics, and named tools matter more than vocal performance.

Can I use AI to help me answer an AI interview?

Using AI to prep beforehand (rehearse PAR answers, gap-analyse the JD) is fine. Using AI live in the call to feed you answers is fraud and is increasingly detectable. Voice agents can flag unnatural pause patterns, mismatched filler-word frequency, and overly polished phrasing. Some platforms now run liveness and audio-authenticity checks. If you get caught, you will not just lose the role; many ATS platforms now share fraud signals across employers.

How long is an AI pre-screen interview, and what should I prepare?

Most AI pre-screens run 10 to 20 minutes and cover 5 to 8 questions. Prepare 4 to 6 PAR (Problem, Action, Result) stories that cover your main competencies with named tools, real metrics, and concrete timeframes. Test your microphone, pick a quiet room, and treat it like a phone screen with a human: be conversational, take a breath before answering, and ask the agent to repeat questions if needed (most modern agents handle that fine).

Want your real career surfaced into PAR-ready answers for every AI pre-screen? Join Hirelytica and turn your full experience into a structured library.

📊 Sources & Research

🔬 Industry Reports

Greenhouse 2026 Candidate AI Interview Report: 2,950 active job seekers; 63% interviewed by AI, 70% not disclosed, 38% have withdrawn (greenhouse.com)
SHRM Talent Trends 2025: 67% comfortable with AI initial screening if humans make final decision; 79% want disclosure
Aisera 2026 AI Recruiting Guide: 46% of orgs adopting agentic AI in TA; 80% of high-volume roles to start with AI voice screen by mid-2026
Gartner (Oct 2025): 75% of hiring processes by 2027 will include AI proficiency tests; 55% of supply chain leaders expect agentic AI to reduce entry-level hiring needs (gartner.com)

📈 Platform & Field Data

Greenhouse Acquires Ezra AI Labs (2026): Conversational AI to be embedded directly in Greenhouse ATS (greenhouse.com)
Paradox / Olivia case studies: 100+ simultaneous candidate conversations; 5-7 day workflows compressed to under 48 hours (paradox.ai)
HeyMilo / Carv platform data: 70%+ AI phone-screen completion rate vs 42% video dropout; 53% pass rate at next round vs 29% resume-only (heymilo.ai, carv.com)
Jabarian & Henkel (2025) Field Experiment: AI vs Human Interviewer offer-acceptance rates statistically indistinguishable (92.14% vs 93.64% across 6,319 offers)
InCruiter Deepfake Detection (2026): Fraudulent activity in 25-30% of flagged sessions, ~2x what human interviewers caught

🔍 Methodology: Synthesis of the Greenhouse 2026 Candidate AI Interview Report (n=2,950), SHRM Talent Trends 2025 survey, Gartner press releases (October 2025 / February 2026), Aisera and Intervuebox 2026 industry analyses, and platform-published completion and pass-rate data from Paradox, HeyMilo, Carv, Sapia, and Mercor. Field-experiment offer-acceptance figures from the Jabarian & Henkel (2025) automated-interview study.