ConnectDevs

Tag: candidate screening

  • Fake Job Candidates: How to Spot and Stop Them in 2026

    Fake Job Candidates: How to Spot and Stop Them in 2026

    Fake Job Candidates: How to Spot and Stop Them in 2026

    Fake job candidates are applicants who invent an identity, borrow someone else’s, or use deepfake video or a stand-in to pass interviews. A few years ago this was a niche security worry. Now it is a screening problem for any team that hires remotely, especially for engineering roles with access to code and customer data.

    This guide covers what the research says about how common candidate fraud is, the warning signs in resumes and video calls, and a layered verification process you can add without turning every interview into an interrogation.

    TL;DR

    • Gartner’s survey of 3,000 candidates found 6% admitted to interview fraud. Gartner also forecasts that by 2028, one in four candidate profiles worldwide will be fake.
    • Palo Alto Networks reported that a researcher with no image-manipulation experience can build a deepfake candidate in as little as 70 minutes.
    • No single check catches fraud. Layer identity checks, consistency checks and at least one live, unscripted conversation.
    • Set rules for acceptable AI use up front, so honest candidates know what is allowed.
    • Keep checks proportionate and consistent, to avoid rejecting real candidates.

    Table of Contents

    How common is candidate fraud?

    Hard numbers are limited, and the best ones come from a few sources. In a Gartner release dated July 31, 2025, 6% of 3,000 surveyed candidates admitted to interview fraud, meaning they posed as someone else or had someone else pose as them. The same release reports that 39% of candidates used AI somewhere in the application process, mostly to write resume text and cover letters.

    Gartner also makes a forecast: by 2028, one in four candidate profiles worldwide will be fake. That is a prediction, not a measurement, so treat it as a signal of direction rather than a current rate.

    On the technical side, HR Dive reported in April 2025 that Palo Alto Networks found a person with no image-manipulation experience could create a fake candidate who passes a video interview in as little as 70 minutes. The same article notes that deepfakes have been used to get malicious actors hired, including people working for North Korea, and that real-time tools let one operator interview for the same job under several identities.

    Types of fake candidates

    Not every case is a deepfake. Most teams will meet four patterns:

    1. Fabricated identity. Invented name, invented work history, often built with AI-written text.
    2. Stolen identity. A real person’s profile and credentials used without their knowledge.
    3. Proxy interviewer. A skilled person takes the interview while someone else is hired and does the work.
    4. Synthetic video or voice. Face or voice altered in real time during a call.

    The first is the most common and the cheapest to attempt. It overlaps with the problem we cover in our guide to screening high-volume applicants, where polished resumes no longer prove anything.

    Warning signs

    In the application

    • A profile created recently, with few connections and activity that starts abruptly.
    • Work history that cannot be traced: companies with no web presence, or titles that do not match the company’s size.
    • A resume tuned to your job post almost word for word.
    • Reference contacts who only reply by chat or email, never by phone, or who share an email domain with the candidate.
    • A photo that appears elsewhere under a different name.

    In the video interview

    Palo Alto Networks, as reported by HR Dive, suggests watching for these:

    • Delays between lip movement and speech.
    • Sudden lighting changes or unnatural edges around the face.
    • Rapid head movements that glitch the image.
    • Refusal or hesitation when asked to turn their head or pass a hand across their face.

    In the answers

    • Fluent, generic answers that fall apart when you ask for a specific detail: which version, which bug, which decision.
    • Long pauses followed by perfect, structured replies, which may mean someone off-camera is feeding answers.
    • Skills that differ sharply between the screen and the technical round.

    Any one of these can have an innocent explanation. A poor connection also causes lag. Use signs to trigger a closer look, not an automatic rejection.

    A layered verification process

    Gartner’s recommended steps are to set expectations about AI use, communicate that you check for fraud, use layered assessments including in-person interviews, and validate identity after hire. Here is how to turn that into a process.

    Stage Check Cost to candidate
    Application Cross-check profile age, employers and links to public work such as code or talks None
    First screen Camera on, ask role-specific follow-ups about past projects Low
    Technical round Live, unscripted problem discussion; ask them to explain trade-offs in their own work Medium
    Final round Different interviewer, ideally in person or with a live identity check against government ID Medium
    Offer Background and reference checks through a verified provider Low
    First 30 days Device and access monitoring, anomaly alerts, identity confirmed on day one None

    Gartner’s release reports that 62% of candidates said they are more likely to apply when the organization requires in-person interviews, so a final in-person or ID-verified step does not necessarily scare off good people.

    Compare what the same candidate says across rounds. Consistent detail across a screen, a technical interview and a reference call is hard to fake. For role-specific interview design, see our guide to reducing time to hire for technical roles, which covers structured interviews without adding delay.

    Staying fair to real candidates

    Over-correcting causes its own harm. A candidate with a bad webcam, a strong accent or an unusual career path is not a fraud risk. Three rules help:

    • Apply the same checks to everyone at the same stage, so you do not single people out.
    • Tell candidates what you check and why. Gartner recommends being open about fraud detection.
    • Define acceptable AI use. Many honest candidates use AI to polish a resume. Say what you allow, such as editing a resume, and what you do not, such as live answer generation.

    If you use automated tools to score or filter candidates, check the legal rules that apply where you hire. Our overview of AI candidate screening software includes a compliance checklist.

    Where AI interviews fit

    An AI interviewer does not replace identity verification, and vendors should say so. The ConnectDevs AI interviewer runs voice interviews with role-specific questions and follow-ups, and produces structured transcripts and ranked evaluations. Because every candidate gets the same structure, you can compare answers across rounds and notice when a story changes.

    ConnectDevs states that it does not claim production detection of rehearsed or AI-generated answers, and that teams should run their own authenticity checks where needed. That is the right expectation for any tool in this category. Use AI interviews to standardize and save time, and keep identity checks as a separate step. For how the main tools differ, read our guide to the best AI interview software, and for the human side, AI recruiter vs human recruiter.

    FAQ

    What are fake job candidates?

    They are applicants who use invented or stolen identities, proxy interviewers or synthetic video and voice to get hired. Motives range from taking a salary to gaining access to company data.

    How common is interview fraud?

    In a Gartner survey of 3,000 candidates, 6% admitted to interview fraud. Gartner forecasts that one in four candidate profiles worldwide will be fake by 2028.

    How can I tell if a candidate is using a deepfake?

    Watch for lip-sync delays, lighting shifts, glitches during head movement and hesitation when asked to turn or wave a hand in front of the face. Confirm with a second live check.

    Can AI detect fake candidates?

    Some identity-verification products can check documents and liveness. General AI interview or screening tools usually do not claim to detect fraud, so verify claims before you rely on them.

    Should I require in-person interviews?

    A final in-person or ID-verified step is the strongest single check. Gartner’s data suggests it may not hurt your applicant flow, since 62% of candidates said they are more likely to apply when in-person interviews are required.

    Is it legal to verify a candidate’s identity?

    Generally yes, but rules on ID checks, biometric data and automated decisions vary by location. Check local law and consult counsel before you add steps.

    Conclusion

    Candidate fraud is real, but the counts are still small and the forecasts are projections. The practical response is a layered process: cheap checks early, a live conversation in the middle and an identity step before the offer. Be open with candidates about what you check, apply the same steps to everyone and review the evidence as a whole before you decide.

    Need consistent first-round interviews you can compare across candidates? Run an AI interview and add your own identity checks on top.

  • How to Screen AI-Generated Resumes Without Rejecting Good Candidates

    How to Screen AI-Generated Resumes Without Rejecting Good Candidates

    How to Screen AI-Generated Resumes Without Rejecting Good Candidates

    AI-generated resumes have made every application look polished, and that has broken the old shortcut of judging candidates by how well they write. Recruiters now face more applications that read the same, a small share of outright fraud, and a real risk of rejecting honest people for using a writing tool.

    This guide explains what the data says, why AI-text detectors are the wrong fix, and a five-check method that screens for evidence instead of prose.

    TL;DR

    • Using AI to polish a resume is common and is not fraud. Gartner found 39% of candidates used AI in their applications.
    • Do not rely on AI-text detectors. They produce false positives and punish honest candidates.
    • Screen for verifiable evidence: specific projects, consistent timelines, and live explanations of past work.
    • Add identity and consistency checks for remote roles, where fraud risk is higher.
    • Use structured conversations early so polish stops being an advantage.

    Table of Contents

    What the data says

    AI use in applications is widespread, but outright fraud is much rarer. Separating the two is the first step.

    According to a Gartner survey:

    • 39% of 3,290 candidates used AI during the application process. Of those, 54% used it for resume text and 50% for cover letter text.
    • 6% of 3,000 candidates admitted to interview fraud, such as impersonating someone or having someone impersonate them.
    • Gartner predicts that by 2028, one in four candidate profiles worldwide will be fake.

    An SHRM article on fraudulent candidates adds that applicant tracking systems mainly screen for skills, experience and keywords, so they are not built to detect deepfakes or fabricated profiles. High volume also pushes recruiters to accept information without checking it.

    So there are two different problems. Most applicants simply used a tool to write better. A small minority are misrepresenting who they are or what they have done. Your process should handle both without treating everyone as a suspect.

    Why AI detectors do not work

    It is tempting to run every resume through an AI-text detector. Resist that. Detectors guess from writing style, and good human writing and edited AI writing look alike. OpenAI withdrew its own AI text classifier because of its low accuracy.

    There is also a fairness problem. Non-native English speakers and candidates who use editing tools for accessibility can be flagged unfairly. And if a tool scores or ranks candidates, it may fall under hiring-automation rules. See our guide to evaluating AI candidate screening software for the compliance questions to ask.

    The five-check method

    Instead of asking “was this written by AI?”, ask “can this person show the work?” These five checks do that.

    1. Look for specifics you can verify

    Generic bullets such as “improved performance and collaborated across teams” say nothing. Look for named systems, measurable outcomes, links to repositories or products, and team sizes. For technical roles, public code and projects are strong evidence. Our guide to finding candidates beyond LinkedIn shows where those signals live.

    2. Check consistency across sources

    Compare the resume with the public profile, the application form and what the candidate says on a call. Mismatched titles, dates or employers are a better signal than writing style. Candidate enrichment can fill in and cross-check profile data automatically.

    3. Hold a short live conversation early

    A 10 to 15 minute conversation about one project, such as “what was the hardest decision, and what would you change?”, is hard to fake and easy to compare. A structured AI interview can run the same questions with every applicant and produce a report, which keeps the process consistent when volume is high.

    4. Use a small work sample for finalists

    Give shortlisted candidates a short, role-relevant task and ask them to explain their approach live. The explanation reveals understanding that polished text cannot.

    5. Verify identity before the offer

    For remote hires, confirm identity and work history with reliable checks. SHRM recommends verifying employment and education directly, noting inconsistencies between resume and interview answers, and using in-person interviews where practical. Gartner’s data supports this, since 62% of candidates say they are more likely to apply to a job that requires in-person interviews.

    Signals that deserve a closer look

    Signal Why it matters What to do
    Vague achievements with no named systems May be generic or fabricated Ask for one detailed example on a call
    Employment dates that overlap or do not match public profiles Inconsistency Verify with the employer
    Candidate cannot explain their own project Work may not be theirs Run a live explanation or code walk-through
    Camera or audio oddities, refusal to do any live step Possible impersonation Add an identity check and a second interviewer
    Hundreds of near-identical applications Mass-apply automation Add a short screening question that needs a specific answer

    A signal is a reason to look, not a reason to reject. Always give the candidate a chance to explain.

    Where tools help

    You cannot read 500 applications carefully, so use tools for the repetitive part and keep people in charge of decisions. For the high-volume playbook, read how to screen high-volume job applicants.

    Good tooling does three things: it compares candidates against the same criteria, it pulls evidence from outside the resume, and it records why a candidate advanced. Whatever you use, keep a human reviewing rejections and keep records of how candidates were evaluated.

    FAQ

    Should I reject candidates who use AI to write their resume?

    No. Many candidates use AI to edit and format, much like spell-check. Judge the evidence behind the claims, not the tool used to write them.

    Can you detect an AI-generated resume?

    Not reliably. Detectors are inaccurate and can flag honest candidates. It is more effective to verify the content through specifics, consistency checks and a live conversation.

    How common is candidate fraud?

    Gartner’s survey found 6% of candidates admitted to interview fraud, and it predicts one in four candidate profiles could be fake by 2028. Rates vary by role and channel, and remote roles face more risk.

    What is the best way to screen high volumes of applications?

    Use consistent criteria, a short structured first conversation and automation for repetitive steps, then review finalists manually.

    How do I spot a deepfake candidate in a video interview?

    Look for visual or audio glitches, delayed responses and reluctance to do spontaneous tasks. Add identity verification, a second interviewer and, where practical, an in-person step before the offer.

    Does this affect hiring-law compliance?

    Tools that score or rank candidates may be regulated in some places, for example New York City’s Local Law 144. Ask vendors about audits, keep human oversight and check with counsel. This is not legal advice.

    Conclusion

    You cannot out-detect AI-written resumes, but you can out-verify them. Screen for evidence, move live conversations earlier and reserve identity checks for finalists. That protects you from fraud without penalizing honest candidates.

    Want every applicant interviewed the same way? Run an AI interview and compare structured reports side by side.

  • How to Screen High-Volume Job Applicants Without Losing Great Candidates

    How to Screen High-Volume Job Applicants Without Losing Great Candidates

    You post one mid-level engineering role on a Friday. By Monday morning you have 340 applications sitting in your ATS, a hiring manager asking for a shortlist by Wednesday, and three other open reqs that need the same treatment. This is the reality most recruiting teams face now, and it’s why so many are searching for a better answer to how to screen job applicants without either burning out or waving through candidates nobody actually vetted.

    This guide walks through a practical process for screening high volumes of applicants without losing the strong candidates buried on page four of your ATS. You’ll get a step-by-step framework, a look at where manual screening breaks down versus where automation genuinely helps, real data on the scale of the problem, and answers to the questions recruiters ask most when they’re rebuilding their screening process.

    TL;DR

    • Applicant volume per job posting has nearly doubled in five years, and most recruiters still spend under two minutes per resume — a mismatch that guarantees good candidates get missed.
    • Speed and quality aren’t opposites if you screen on structured criteria instead of gut feel or keyword matching alone.
    • Automated first-pass screening should filter for stated, job-relevant criteria — not act as an unchecked black box, since research shows AI-driven screening can carry real bias.
    • The highest-leverage fix is usually earlier in the funnel: better sourcing and enrichment reduce how much junk you have to screen out in the first place.
    • A five-step framework (clear criteria, structured questions, checked automation, enrichment, fast interview handoff) scales without sacrificing candidate quality.

    Why High-Volume Screening Breaks Down

    The math has changed. According to BambooHR’s 2026 State of Hiring report, the average number of applicants per job posting climbed from roughly 46 in 2021 to about 95 in 2025 — nearly double in five years, even as the number of completed hires per posting fell. Recruiting teams are doing more screening work to produce fewer hires, not more.

    At the same time, the amount of attention any single resume gets hasn’t grown to match. Survey data from Zety’s Recruiting Preferences Report found that 72% of recruiters spend two minutes or less reviewing a resume before deciding whether to move a candidate forward. That’s not a knock on recruiters — it’s simple math. If your req gets 300 applicants and you have a few hours between interviews and hiring manager syncs to screen them, two minutes each is already generous.

    The result is a system built for speed that quietly punishes candidates whose resumes don’t scream their qualifications in the first ten seconds: career changers, people who undersell themselves, candidates with nontraditional backgrounds, and anyone whose best evidence sits below the fold. If your process only rewards the loudest resume, you’re optimizing for confidence, not competence.

    The Real Cost of Screening Too Fast (or Too Blindly)

    Speed pressure pushes teams toward shortcuts, and one common shortcut — leaning entirely on unchecked automated resume filters — introduces its own risk. A 2025 study by University of Washington researchers Kyra Wilson and Aylin Caliskan, covered by Brookings, tested how large language models rank otherwise-identical resumes when only the associated name changes. White-associated names were preferred over Black-associated names in 85.1% of comparisons where a preference existed, and male-associated names were favored over female-associated names in 51.9% of cases versus 11.1% for the reverse. Equal treatment — no measurable preference either way — happened in a minority of tests.

    That’s not an argument against automation. It’s an argument against automation you can’t inspect. A screening tool that filters on job-relevant, stated criteria (years of experience with a specific tool, licensure, location eligibility) behaves very differently from one making opaque judgment calls based on patterns in a training set. The fix isn’t “go back to manual review” — manual review has its own well-documented bias problems and it’s the reason time-to-fill keeps climbing while quality stays flat. The fix is screening logic you can see, test, and defend.

    This is also where the cost of getting it wrong compounds. A candidate incorrectly screened out never shows up in your funnel data as a mistake — you just don’t know what you didn’t see. A candidate incorrectly screened in costs you far more visibly, which is a big part of why the true cost of a bad hire runs so much higher than most budgets assume.

    A Practical Framework for Screening Hundreds of Applicants

    Here’s the process that holds up whether you’re screening 50 applicants or 500.

    1. Write pass/fail criteria before you open a single resume

    Before the applications start rolling in, get the hiring manager to name the three to five things that are genuinely non-negotiable for the role — not “nice to have,” actually non-negotiable. For a compliance analyst role that might be “active CPA license” and “2+ years in a regulated industry.” For a support engineer role it might be “has debugged production incidents” and “available for the on-call rotation.” Everything else is a ranking factor, not a gate. Skipping this step is why screening turns into a debate about each individual resume instead of a repeatable filter.

    2. Replace gut-check reading with structured screening questions

    Instead of reading full resumes top to bottom for every applicant, use a short set of screening questions (in your ATS, a form, or an initial async interview) tied directly to your pass/fail criteria. This does two things: it standardizes what every candidate is judged against, and it surfaces disqualifying or qualifying information faster than resume archaeology does. It’s the same logic behind why traditional pass/fail assessments frustrate strong candidates — the goal is targeted signal, not a hoop to jump through.

    3. Automate the first pass, but audit what it’s actually optimizing for

    Automated screening earns its place here: it’s the only way to apply your criteria consistently across hundreds of applicants without recruiter fatigue creeping in by application 150. The catch is that “automated” isn’t automatically fair or accurate. Before you trust a tool with your first pass, know what it’s scoring on. Matching based on actual skills, experience, and stated intent performs very differently than matching based on keyword overlap with the job description — which is part of why intent-based AI matching outperforms keyword search for this exact reason. Keyword matching rewards resume-writing skill; intent and skills matching rewards actual fit.

    4. Enrich thin resumes before you reject them

    A short resume isn’t the same as a weak candidate — it’s often just a candidate who didn’t over-optimize their document for an ATS. Before you screen someone out for “insufficient detail,” it’s worth pulling in outside signal (portfolio work, GitHub activity, verified work history, public project contributions) rather than rejecting on a thin PDF alone. This is the specific gap that AI candidate enrichment closes — it builds out a fuller picture so your screen isn’t just judging someone’s resume-writing.

    5. Move your shortlist into structured interviews fast

    Screening isn’t the finish line — it’s supposed to produce a short list you can move on quickly. If your best candidates from Monday’s screen aren’t in a scheduled conversation by Wednesday, a competing offer usually gets there first. Once you’ve narrowed the pool, treat scheduling speed as part of the screening process, not a separate problem to solve later. That urgency is the whole idea behind the 24-hour rule for interview scheduling.

    Manual vs. Automated Screening: What Actually Scales

    Neither approach is universally right. Here’s how they actually compare once volume climbs past what one recruiter can reasonably read.

    Factor Manual screening Automated / AI-assisted screening
    Consistency across candidates Degrades as reviewer fatigue sets in (candidate #200 gets less attention than candidate #5) Applies the same criteria to every applicant, every time
    Speed at volume Roughly two minutes per resume, best case — untenable past a few hundred applicants Screens hundreds of applicants in the time it takes to read a handful manually
    Bias risk Documented human bias toward familiar schools, names, and resume formatting Can reduce human bias if criteria are transparent and job-relevant; can encode new bias if the model is a black box
    Nuance and context Better at catching non-obvious fit signals a form can’t capture Needs enrichment or human review layered on top to catch context a resume alone misses
    Best use case Final-round judgment calls, small applicant pools, senior or highly specialized roles First-pass filtering at volume, freeing recruiter time for judgment calls that actually need a human

    The teams that scale well don’t pick one column — they use automation for the first-pass filter and route the shortlist to a human for the judgment calls that actually need one. That’s also the core case for recruitment automation generally: it’s not about removing recruiters from the process, it’s about removing the part of the process that shouldn’t need a recruiter’s full attention in the first place.

    Common Mistakes That Cost You Good Candidates

    • Screening on keywords instead of substance. A resume that mirrors the job description word-for-word isn’t automatically the best-fit candidate — it might just be the most keyword-optimized one.
    • Letting the screen do the interview’s job. If you’re trying to fully assess someone in a resume scan, you’ll either reject too aggressively or advance people you haven’t actually vetted. Save deep evaluation for structured interviews, where you can actually probe.
    • No documented criteria, so every recruiter screens differently. Without a shared rubric, two recruiters screening the same applicant pool will produce two different shortlists.
    • Confusing “screened quickly” with “screened well.” Passing a candidate through a 30-minute phone screen that’s really a soft skills chat isn’t the same as verifying they meet your actual criteria — it’s often just 10 hours of your week disappearing into calls that don’t move the decision forward.
    • Treating charisma as a proxy for competence. Confident self-presentation and job performance are not the same thing, which is a big reason so many bad hires make it through the interview stage in the first place — the screening criteria that let them through were measuring the wrong thing.

    FAQ

    How do I screen hundreds of applicants without missing good candidates?

    Set pass/fail criteria before applications arrive, use structured screening questions tied to those criteria, and let automated first-pass filtering apply them consistently across every applicant. Reserve manual judgment for the shortlist that clears the first pass, where nuance actually matters.

    What is the best resume screening software for high-volume hiring?

    The best option is whichever tool screens on job-relevant skills and stated intent rather than raw keyword density, and lets you see and adjust the criteria it’s filtering on. A tool you can’t inspect or tune isn’t actually saving you time — it’s just moving the risk of a bad filter somewhere you can’t see it.

    Is AI resume screening biased?

    It can be, depending on what the model is trained on and what it’s optimizing for. Research from the University of Washington covered by Brookings found significant race and gender disparities in how large language models ranked otherwise-identical resumes. The safeguard is using tools that filter on transparent, job-relevant criteria rather than opaque pattern matching, and auditing outcomes periodically.

    How long should resume screening take per candidate?

    Most recruiters currently spend under two minutes per resume, according to Zety’s Recruiting Preferences Report, which is part of why so many strong but understated resumes get missed. The goal isn’t to spend longer per resume — it’s to spend that limited time on criteria that actually predict job performance, with automation handling the pure pass/fail filtering.

    Should I use phone screens or async screening for high applicant volume?

    Async, structured screening questions scale better for the top of a large funnel because they don’t require scheduling and can be reviewed in batches. Save live phone or video screens for candidates who’ve already cleared your baseline criteria, so you’re not spending scheduling time on candidates who wouldn’t have passed anyway.

    How do I reduce bias in candidate screening?

    Screen against documented, job-relevant criteria instead of subjective impressions, standardize the questions every candidate answers, and if you use automated screening, understand and periodically check what signals it’s actually weighting. Removing names and schools from an initial screen can also help, but it only works if the underlying criteria are sound to begin with.

    What’s the difference between screening and sourcing?

    Sourcing is finding and attracting candidates into your pipeline; screening is evaluating the candidates already in it against your criteria. Better sourcing — including AI candidate sourcing that targets people who actually match the role — reduces how much low-fit volume you have to screen out later, which is often a bigger time-saver than optimizing the screen itself.

    Where to Go From Here

    Screening high volumes of applicants well comes down to a specific trade you have to make deliberately: decide what’s a non-negotiable gate versus a ranking factor, apply that consistently with automation you can actually inspect, and save human judgment for the shortlist where it earns its cost in time. Teams that skip the “decide the criteria first” step end up either drowning in applications or fast-screening their way past candidates who deserved a second look.

    If your team is still screening applicants one resume at a time while volume keeps climbing, it’s worth looking at where the bottleneck actually starts. Often it’s not the screen itself — it’s an applicant pool full of people who were never a strong match to begin with. Explore AI candidate sourcing with ConnectDevs to see what it looks like when the people entering your funnel are already closer to what you’re looking for.