ConnectDevs

Tag: AI hiring compliance

  • AI Hiring Laws in 2026: What Recruiters Need to Know Before Using AI Tools

    AI Hiring Laws in 2026: What Recruiters Need to Know Before Using AI Tools

    AI Hiring Laws in 2026: What Recruiters Need to Know Before Using AI Tools

    AI hiring laws now apply to ordinary recruiting work such as resume ranking, candidate scoring and automated interviews. The rules differ by location, several deadlines moved during 2026, and many online summaries are already out of date.

    This guide summarizes what is in force and what is coming, based on the sources linked below as of October 2026. It is a plain-English overview for recruiters and founders, not legal advice. Always confirm details with counsel.

    TL;DR

    • NYC Local Law 144 has been enforced since July 5, 2023 and requires bias audits and candidate notice for automated employment decision tools.
    • Illinois (HB 3773) and Texas (TRAIGA) took effect January 1, 2026. California’s automated-decision regulations took effect October 1, 2025.
    • Colorado’s rewritten AI law applies to decisions made on or after January 1, 2027.
    • The EU’s high-risk hiring obligations have been postponed to December 2, 2027, though many summaries still say August 2026.
    • Practical steps help in every jurisdiction: inventory tools, give notice, keep humans in the loop, keep records and ask vendors for audit evidence.

    Table of Contents

    Snapshot: what applies where

    The table below summarizes the main rules. Sources are secondary trackers, so verify dates against the official text before relying on them.

    Jurisdiction Law Status Main duty
    New York City Local Law 144 Enforced since July 5, 2023 Bias audit, public summary, candidate notice
    Illinois HB 3773 Effective January 1, 2026 No discriminatory AI use; notice to applicants
    California Civil Rights Council ADS regulations Effective October 1, 2025 Anti-discrimination rules; four-year records
    Texas TRAIGA (HB 149) Effective January 1, 2026 No AI built or used with intent to discriminate
    Colorado AI Act as rewritten by SB 26-189 Applies to decisions on or after January 1, 2027 Disclosure and consumer rights
    European Union AI Act, high-risk systems Obligations from December 2, 2027 Risk management, transparency, human oversight

    New York City Local Law 144

    Local Law 144 requires employers in New York City to audit automated employment decision tools for bias and to tell candidates about their use. According to the NYC Department of Consumer and Worker Protection, the key requirements are:

    • The tool must have had an independent bias audit within one year of its use.
    • Information about the audit must be publicly available.
    • Candidates must receive notice, and the department has clarified that this must be at least 10 business days before the tool is used.

    Penalties are $500 for a first violation and $500 to $1,500 for each later violation, and each day of use and each missed notice can count separately, according to a Metaview summary of 2026 and 2027 rules. A study cited in HeroHunt’s guide found that only 18 of 391 covered employers had posted the required audit, so awareness remains low.

    Illinois, California, Texas and Colorado

    Illinois

    HB 3773 amends the Illinois Human Rights Act. As of January 1, 2026, employers may not use AI in ways that discriminate against protected classes, may not use ZIP codes as a proxy for protected classes, and must notify applicants and employees that AI is used. Enforcement runs through the existing civil-rights complaint process.

    California

    The Civil Rights Council’s automated-decision regulations, effective October 1, 2025, apply to employers with five or more employees. Liability can extend to vendors and consultants acting as agents. Employers must keep records of the system, its selection criteria and its outputs for four years, and anti-bias testing evidence is relevant to a defense.

    Texas

    TRAIGA took effect January 1, 2026. It prohibits developing or deploying AI with the intent to unlawfully discriminate, a narrower standard than outcome-based tests. The Texas Attorney General has exclusive enforcement, there is no private right of action and a 60-day cure period applies.

    Colorado

    Colorado’s original AI Act was delayed more than once. According to Metaview and HeroHunt’s state-by-state guide, SB 26-189, signed in May 2026, rewrote the law. It removed the duty to prevent algorithmic discrimination, the mandatory impact assessments and the deployer risk-management program, and replaced them with a disclosure-and-rights model enforced by the Attorney General. Its duties apply to consequential decisions made on or after January 1, 2027. A federal court order in April 2026 also limits enforcement for a period tied to a pending ruling, so watch for updates.

    The EU AI Act

    The EU AI Act treats AI used in recruitment and candidate evaluation as high-risk. Many articles still say the high-risk obligations begin on August 2, 2026. According to Metaview, Regulation (EU) 2026/1744, adopted July 8, 2026 and in force since July 27, 2026, moved the start date for high-risk hiring obligations to December 2, 2027.

    Systems already on the market before that date are covered only if their design is subject to significant changes afterward. Penalties for breaching the Act’s prohibited practices can reach 7% of worldwide annual turnover or €35 million, whichever is higher. This is a single secondary source for the new date, so confirm it against the Official Journal before you plan around it.

    If you recruit in the EU, the practical impact is on vendors: expect providers to offer risk documentation, transparency information and human-oversight features before the deadline.

    A practical compliance checklist

    1. Inventory every AI tool in your hiring process. Include sourcing rankings, resume screening, chatbots, interview scoring and assessments.
    2. Decide where decisions are automated. A tool that only suggests is different from one that rejects candidates. See our guide to AI candidate screening software.
    3. Give notice. Tell candidates when AI is used, in clear language and early enough to meet local rules.
    4. Keep a human in the loop. Have a person review rejections and be able to override outputs.
    5. Request bias-audit evidence. Ask vendors for audit results or documentation, and get them in writing.
    6. Keep records. Store criteria, outputs and decisions. California requires four years for covered systems.
    7. Offer accommodations. Give candidates a way to request an alternative to automated steps.
    8. Review annually. Laws and deadlines are still changing, and this article will be out of date in months.

    For a broader view of what to automate and what to leave to people, see what AI recruiting agents actually automate.

    Questions for AI recruiting vendors

    • Does the product score, rank or reject candidates, or only organize information?
    • Do you have an independent bias audit, and can we see the summary?
    • What data does the tool use, and does it infer protected characteristics?
    • Can recruiters see and override the reasoning?
    • How long do you store candidate data, and how can candidates opt out?
    • Who is responsible if a regulator asks for records?

    Evaluation matters at the interview stage too. Our guide to AI interview software covers what structured reporting should include. ConnectDevs’ AI interviewer produces a structured report for each interview that a recruiter reviews, but it does not replace your own compliance review of how you use it.

    FAQ

    Is it legal to use AI in hiring?

    Generally yes, but a growing set of laws governs how. Requirements include notice, bias audits, records and human oversight, depending on where candidates and employers are located.

    What is NYC Local Law 144?

    It is a New York City law, enforced since July 5, 2023, that requires employers using automated employment decision tools to have an independent bias audit, publish a summary and give candidates notice.

    When does the EU AI Act apply to hiring?

    Reports indicate high-risk hiring obligations now begin on December 2, 2027, after a 2026 regulation postponed the earlier August 2026 date. Confirm the current date with official sources.

    Do these laws apply to small companies?

    Some do and some do not. California’s regulations apply to employers with five or more employees, while NYC’s law applies to covered employers and agencies using tools for NYC candidates. Check each law’s scope.

    Do I have to tell candidates I use AI?

    In several places yes, including NYC and Illinois. Even where it is not required, clear notice builds trust. Gartner found only 26% of candidates trust AI to evaluate them fairly.

    Who is liable, the employer or the vendor?

    Usually the employer remains responsible, and in California liability can also extend to vendors acting as agents. Ask vendors for contractual commitments and documentation.

    Conclusion

    You do not need to avoid AI to stay compliant. You need to know which tools influence decisions, tell candidates, keep people in charge and keep records. Start with an inventory this month, because the rules will keep moving.

    Want to see how an AI interview report works in practice? Book a demo and walk through it with your own role.

  • Recruiting MCP: What It Is and 8 Questions to Ask Before You Connect AI to Your Hiring Stack

    Recruiting MCP: What It Is and 8 Questions to Ask Before You Connect AI to Your Hiring Stack

    Recruiting MCP: What It Is and 8 Questions to Ask Before You Connect AI to Your Hiring Stack

    Recruiting MCP is the use of the Model Context Protocol to let an AI assistant such as Claude or ChatGPT read from and act inside your recruiting tools. Instead of opening a sourcing platform, you type a request in plain language, and the assistant calls the tool for you. A few vendors already offer it, and the idea is moving fast. It also raises real questions about permissions, audit trails and hiring law.

    This guide explains what recruiting MCP is, what vendors expose today and eight questions to ask before you connect an assistant to candidate data. It is written for recruiters, talent leaders and founders who are hearing the term and need to separate useful from risky.

    TL;DR

    • MCP is an open standard for connecting AI applications to outside tools and data. The official site compares it to a USB-C port for AI.
    • SeekOut, Get on Board, Findem and others already expose recruiting data through MCP, and vendors differ sharply on what an assistant is allowed to do.
    • An assistant inherits the permissions of the person who connects it. Start read-only and keep a human on every decision.
    • Ask vendors about scopes, contact data, write actions, logging and revocation before you connect anything.

    Table of Contents

    What is recruiting MCP?

    Recruiting MCP is an MCP server that a recruiting vendor runs so that AI assistants can search candidates, read job data and take approved actions through your account. The Model Context Protocol itself is an open-source standard for connecting AI applications to external systems, including data sources, tools and workflows. Its documentation lists Claude and ChatGPT among supported clients.

    Think of the server as a doorway. Get on Board, a Latin American tech job board, describes it in its recruiter explainer as a way for the AI tool you already use to connect to an outside service, so you ask in plain language and the tool uses your account to answer. MCP does not make recruiting data accurate or lawful on its own. The vendor still has to enforce identity, permissions and audit rules.

    What vendors offer today

    Offerings differ a lot. This table summarizes what each source states publicly.

    Vendor What an assistant can do Limits stated
    SeekOut Analyze talent markets, build ranked shortlists, draft outreach for review, compare candidates, retrieve contacts, export and push to an ATS Requires a SeekOut Recruit license; browser-based OAuth sign-in
    Get on Board Search talent, view profiles, list jobs and job details No emails, phones or CV files returned; cannot unlock candidates or invite them; connection expires after 2 hours
    Findem Findem Studio offers pre-built or custom agents and MCPs that other AI tools can call Pricing and scope not public
    Workable and Greenhouse Per Hirevire, Workable exposes 94 tools and Greenhouse 36 Counts are from a vendor blog, so verify in each vendor’s docs

    The SeekOut details come from its MCP quickstart, which lists support for Claude, ChatGPT, Codex, Claude Code, Gemini Enterprise and Microsoft 365 Copilot. Hirevire, which sells screening software, reports that eleven HR tools ran their own MCP server as of August 2026, so treat its counts as the author’s own check.

    Note on ConnectDevs: its own Recruiting MCP guide states that ConnectDevs does not currently publish an MCP server. It presents a product-design proposal for read-only search first, then confirmed enrichment and shortlist writes. We say this plainly so you can weigh vendor claims, including ours, against what is live.

    What recruiters can do with it

    The common patterns are narrow and practical:

    • Shortlist from a brief. Ask for ten candidates in a region who match an open role and a summary of the top five. Get on Board gives this exact example.
    • Read the pipeline. Pull candidate lists and statuses without opening a dashboard.
    • Draft, then review. Generate outreach for you to edit and send. Our outreach response rate guide shows why targeting and review matter more than volume.
    • Summarize interviews. Have an assistant condense transcripts, with a human reading the summary.

    These save time on research and admin. They do not replace judgment about who to hire. For where automation helps and where it should stop, read which hiring workflows to automate first and who does what between AI and recruiters.

    Risks to plan for

    The main risk is that an assistant acts with your permissions and can do more than you intend. A connected assistant can read whatever your account can read. Four risks deserve attention.

    1. Over-broad access. Hirevire recommends OAuth, scoped permissions, one-click revocation and human confirmation on any write action.
    2. Automated selection. Hirevire argues an agent should never auto-reject, auto-advance or rank-and-cut candidates without human review. It cites EEOC guidance on employer responsibility for AI-driven selection and NYC Local Law 144’s bias audit rule. Confirm the details with counsel.
    3. Contact data leakage. Some vendors, such as Get on Board, keep emails and phones out of the assistant’s reach. Others return contacts. Know which one you have.
    4. Sensitive data in a chat. Candidate records in a general-purpose assistant raise privacy and retention questions. Check where the data goes and how long it stays.

    Executives want humans in charge. Hirevire cites a 2026 SHRM survey of recruiting executives in which 85% want to keep final decision authority, while 84% plan to use AI in recruiting this year. Our guide to AI candidate screening software covers compliance checks in more detail.

    8 questions to ask a vendor

    1. Which tools does the server expose? Ask for the full list and which are read-only versus write.
    2. What are the permission scopes? Look for separate scopes for search, contact data, writes, outreach and interviews.
    3. Is contact data returned to the assistant? If yes, who can see it and is it logged?
    4. Which actions need confirmation? Costly or candidate-facing actions should show a preview and require a click.
    5. How is access authenticated and revoked? OAuth with one-click revocation is the baseline.
    6. What gets logged? You want a record of who asked what, which candidates were touched and what changed.
    7. How are credits or costs metered? An assistant can run many searches quickly, so ask for limits and previews.
    8. What happens to data after a session? Ask about retention, deletion and whether models train on it.

    A safe rollout in four phases

    1. Read-only search. Let the assistant search and summarize. Nothing is written or sent.
    2. Confirmed writes. Allow saving shortlists and enrichment only after a preview and a click.
    3. Interview initiation. Start structured interviews through the assistant, with consent and notice for candidates.
    4. Scoped outreach. Permit narrow campaign actions last, with sending limits and review.

    The same order works for any vendor. If you are still choosing a sourcing platform, our guide to the best AI recruiting tools explains the categories, and ConnectDevs’ AI sourcing tool shows how a ranked, explained search works in the product today.

    FAQ

    What does MCP stand for in recruiting?

    MCP stands for Model Context Protocol, an open standard for connecting AI applications such as Claude or ChatGPT to external tools and data. In recruiting, it lets an assistant query sourcing, ATS or interview tools through your account.

    Which recruiting tools have an MCP server?

    SeekOut, Get on Board and Findem describe MCP offerings, and Hirevire reports that Workable and Greenhouse expose tools too. Check each vendor’s documentation, because the list changes quickly.

    Is recruiting MCP safe?

    It can be, with scoped permissions, confirmation on write actions, logging and revocation. It is not safe by default, because the assistant inherits your access.

    Can an AI assistant make hiring decisions through MCP?

    It should not. Keep a human on every decision to advance or reject, and confirm the legal duties for automated tools in your jurisdiction.

    Do I need a developer to use recruiting MCP?

    Usually not. SeekOut’s documentation describes browser-based sign-in from a supported assistant. Enterprise platforms may need an administrator to provision the connector first.

    Does ConnectDevs have an MCP server?

    Not currently. Its Recruiting MCP guide says it does not publish one and describes a proposed design. Its Scout sourcing, enrichment, Pilot outreach and AI interviewer are available in the product itself.

    Conclusion

    Recruiting MCP turns your tools into something you can talk to, which saves time on search and admin. It also hands an assistant your permissions. Start read-only, ask the eight questions, keep humans on every decision and expand access one phase at a time.

    Want to see ranked, explained candidate search in action? Explore ConnectDevs AI candidate sourcing and test it on a live role.

  • AI Recruiting Automation: Which Hiring Workflows to Automate First (and Which to Leave Alone)

    AI Recruiting Automation: Which Hiring Workflows to Automate First (and Which to Leave Alone)

    AI Recruiting Automation: Which Hiring Workflows to Automate First (and Which to Leave Alone)

    AI recruiting automation pays off when you automate the right workflows in the right order. Most teams do the opposite. They start with the flashy step, such as AI screening, and skip the boring ones that save the most time with the least risk.

    This guide gives you a four-factor scoring framework, a workflow-by-workflow table, ROI math with labeled assumptions, and the compliance guardrails that matter in 2026. It is built for TA leaders, agency owners and founders who need a sequence, not a tool list.

    TL;DR

    • Score every workflow on volume, repeatability, trust risk and ease of verification. Automate the high scorers first.
    • Scheduling, outreach follow-ups and ATS hygiene come first. Screening and interviews come later, with human review. Offers and closing stay human.
    • In our illustrative model, four low-risk workflows save about 165 recruiter hours a month. Replace the numbers with your own.
    • Regulators treat tools that filter or evaluate candidates as the risky category. Plan audits, notices and human oversight before you switch those on.

    Table of Contents

    What is AI recruiting automation?

    AI recruiting automation is the use of software, including AI models, to run repeatable hiring tasks with little manual effort, while people keep control of decisions that affect candidates.

    It covers a wide range. Some tasks are simple rules, like sending a calendar link. Others use AI, like ranking profiles or running a structured interview. The difference matters, because the risk grows as the task moves from administration to judgment.

    Adoption is already common. SHRM’s 2025 Talent Trends report found that 51% of organizations use AI in recruitment. So the question is no longer whether to automate. It is where to start.

    If you want a deeper look at what agents can and cannot do, read our guide to what AI recruiting agents actually automate in 2026. This post takes a different angle. It ranks workflows by how safe and how valuable they are to automate.

    The four-factor prioritization framework

    Score each workflow from 1 to 5 on four factors: volume, repeatability, trust risk and ease of verification. A higher total means automate sooner.

    The scores are a judgment call, not a law of nature. Run them with your team and argue about the numbers. The argument is the useful part.

    The four factors

    • Volume. How often does the task happen? 5 means hundreds of times a month. 1 means a few times a quarter.
    • Repeatability. Does the task follow the same steps each time? 5 means a fixed pattern. 1 means every case is different.
    • Trust risk (inverted). What happens to candidate trust if the automation fails? 5 means low risk, because a failure is invisible or trivial. 1 means a failure could harm a candidate or your employer brand.
    • Ease of verification. Can a human check the output quickly? 5 means one glance. 1 means you cannot tell if it is wrong without redoing the work.

    Trust risk deserves extra weight. Gartner surveyed job candidates and found that only 26% trust AI to evaluate them fairly, and 25% trust employers less when AI evaluates their information. Candidates accept AI that schedules their interview. They are far more wary of AI that judges them.

    How to read the total

    • 16 to 20: automate now.
    • 12 to 15: automate with human review on every output.
    • Below 12: pilot carefully or keep human-led.

    Workflow-by-workflow table

    Scheduling, outreach follow-ups and ATS hygiene score highest. Screening and interviews score lowest, because trust risk and verification are hard.

    These scores are our own illustrative judgment for a typical mid-size recruiting team. Adjust them to your context.

    Workflow Volume Repeat-ability Trust risk (5 = low) Verification Total /20 Verdict
    Interview scheduling 5 5 5 5 20 Automate now
    Outreach follow-ups 5 4 4 5 18 Automate now
    Reporting and ATS hygiene 4 5 5 4 18 Automate now
    Candidate enrichment 4 4 4 3 15 Automate, spot-check
    Sourcing 5 3 4 3 15 Automate, review shortlists
    Application screening 5 3 2 2 12 Human in the loop
    First-round interviews 4 3 2 2 11 Pilot with oversight
    Final interviews, offers, closing 2 1 1 1 5 Keep human

    The rollout order, in detail

    Start with administrative workflows where a mistake is cheap and obvious, then move toward judgment workflows only after you have measured results.

    1. Interview scheduling

    Scheduling is pure coordination. It has high volume, a fixed pattern and an instant check: the meeting happens or it does not. Automate it first.

    2. Outreach follow-ups

    Most replies come after a follow-up, and most recruiters stop too early because follow-ups are tedious. Automate the sequence, but make sure it stops when a candidate replies. Nothing erodes trust faster than a “just checking in” email sent to someone who already said no.

    Pilot, our outreach product, runs multi-touch outreach from your own Gmail, stops follow-ups on reply and sorts replies. If you are building this yourself, those two behaviors are the minimum bar.

    3. Reporting and ATS hygiene

    Stage updates, notes, duplicate records and weekly reports eat hours and add no candidate risk. Automate the data entry. Keep a person responsible for the numbers leadership sees. For benchmarks on what to track, see our cost per hire benchmark for engineering roles.

    4. Candidate enrichment

    Enrichment fills in missing profile data, such as role history and contact details. It is repeatable and high volume. Verification is harder, since stale data can look right. Spot-check a sample each week. ConnectDevs offers candidate enrichment for this step.

    5. Sourcing

    AI sourcing finds and ranks candidates faster than manual Boolean strings. It is not a decision about a person, because a recruiter still chooses who to contact. Review every shortlist before outreach. Our guide to finding passive candidates beyond LinkedIn covers the channels. For a tool example, Scout runs natural-language search over a published 800M+ public-profile graph and returns ranked matches with reasoning.

    6. Application screening

    Screening is where volume tempts teams to go all in. It is also where bias, error and regulation concentrate. Use AI to organize and summarize, and let a recruiter make the reject decision. Our walkthrough on how to screen high-volume job applicants shows a workable split.

    7. First-round interviews

    Structured AI interviews can standardize first-round questions and produce transcripts. They are also the most visible AI touchpoint for a candidate. Disclose them, offer an alternative where you can, and keep a human reviewing the results. SAM, the ConnectDevs AI interviewer, runs voice-based structured screening interviews with transcripts and ranked reports. To compare options, see our roundup of the best AI interview software in 2026.

    What to leave alone

    Keep humans in charge of final hiring decisions, offer negotiation, closing, and any conversation where a candidate is making a life choice.

    Those moments run on persuasion, empathy and context. They are low volume, so automation saves little. They carry the highest trust cost if you get them wrong. That combination is a clear “leave alone.”

    Also leave alone anything you cannot verify. If you cannot explain why the tool rejected a candidate, do not let it reject candidates.

    Fraud is a related reason to keep humans involved. Gartner predicts that by 2028, one in four candidate profiles worldwide will be fake. Automated pipelines that never meet a human are easier to game.

    ROI math with illustrative assumptions

    Even conservative assumptions show four low-risk workflows saving roughly 165 recruiter hours a month, but the numbers below are illustrative, not benchmarks.

    Every figure here is an assumption chosen to make the math easy to follow. Swap in your own volumes, time estimates and costs. None of this is a ConnectDevs customer result.

    Assumptions

    • Fully loaded recruiter cost: $60 per hour (illustrative).
    • Team of five recruiters.
    • Automation platform costs: $1,500 per month in total (illustrative).
    Workflow Monthly volume (assumed) Manual time per unit (assumed) Manual hours Time cut (assumed) Hours saved
    Interview scheduling 200 interviews 12 min 40 75% 30
    Outreach follow-ups 1,500 prospects x 2 follow-ups 2 min 100 80% 80
    Application screening (admin only) 600 applicants 4 min 40 50% 20
    Reporting and ATS hygiene 5 recruiters x 13 hours n/a 65 50% 32.5
    Total 245 162.5

    The math

    • Hours saved: 162.5, or about 165.
    • Value of time saved: 162.5 x $60 = $9,750 per month.
    • Less tool cost: $9,750 – $1,500 = $8,250 net per month.
    • Annualized: roughly $99,000.

    Two cautions apply. First, saved hours are only worth money if you redirect them to higher-value work, such as talking to candidates. Second, the screening row assumes AI handles organization and humans make the decisions. If you automate the decision itself, the compliance cost changes the math.

    The strongest payoff is usually speed. Faster scheduling and follow-ups shorten the process, which we cover in our guide to reducing time to hire for technical roles. For a view of output per recruiter, see recruiter productivity and placements per recruiter.

    Compliance guardrails

    The more a tool filters, ranks or evaluates candidates, the more regulation applies. Administrative automation faces far less scrutiny.

    This section reflects our reading of public sources as of October 2026. It is not legal advice. Confirm your obligations with counsel.

    New York City Local Law 144

    NYC requires employers and agencies using automated employment decision tools to have a bias audit within one year of using the tool. They must also publish the audit results and give candidates notice at least 10 business days before use. Enforcement began on July 5, 2023.

    Enforcement is still evolving. A New York State Comptroller audit published in December 2025 reviewed DCWP’s enforcement from July 2023 to June 2025 and found gaps, and DCWP committed to implementing most of its recommendations. Expect scrutiny to rise, not fall.

    EU AI Act

    The AI Act classifies systems used to recruit or select people, including to analyse and filter job applications and to evaluate candidates, as high-risk. The original deadline for those obligations was August 2, 2026. The Digital Omnibus has since been published in the Official Journal, and high-risk obligations now apply from December 2, 2027.

    That delay is not a free pass. Rules on prohibited practices and AI literacy already apply, and high-risk compliance takes time to build.

    A practical guardrail checklist

    • Keep a human decision-maker for every reject and advance decision.
    • Disclose AI use to candidates before it touches their data.
    • Get audit evidence from vendors for any tool that screens or ranks.
    • Log outputs so you can explain a decision later.
    • Test for bias on your own pipeline data, not just vendor claims.
    • Set stop rules, such as halting automation when error rates pass a threshold.

    Smaller teams can follow the same logic. Our piece on AI recruiting software for startups covers how to apply it without a big compliance budget.

    FAQ

    What should I automate first in recruiting?

    Automate interview scheduling first. It has high volume, a fixed pattern, low candidate risk and an instant check. Outreach follow-ups and ATS hygiene come next.

    Should I automate candidate screening?

    Automate the organizing, not the deciding. Let AI summarize and rank, and have a recruiter make the final call. Screening tools can trigger bias-audit and notice rules, so check your local law first.

    How do I calculate the ROI of recruiting automation?

    Multiply monthly volume by minutes per task, estimate the share you can cut, and price the saved hours at your loaded recruiter cost. Subtract tool costs. Then check that the saved time goes to higher-value work.

    Does NYC Local Law 144 apply to scheduling or outreach tools?

    The law targets automated employment decision tools, which are tools that support hiring decisions. Pure scheduling and follow-up sequencing generally do not make those decisions. Confirm the scope with counsel for your specific tool.

    When do EU AI Act rules for hiring tools apply?

    High-risk obligations for recruitment and selection systems now apply from December 2, 2027, after the Digital Omnibus moved the original August 2026 date. Prohibited-practice and AI literacy rules already apply.

    Will AI replace recruiters?

    Not for the work that matters most. AI absorbs repetitive tasks. Recruiters keep relationships, judgment and closing. Our guide to what AI recruiting agents still cannot do goes deeper.

    Conclusion

    Good AI recruiting automation is a sequencing problem. Score your workflows, start with the high-volume, low-risk ones, and add judgment workflows only after you can measure and verify them. Keep people in charge of decisions and closing.

    Run the scoring exercise with your team this week. Then pick the top two workflows and set a 30-day target for each.

    Want to see how this fits your pipeline? Book a demo and we will walk through sourcing, enrichment, outreach and structured interviews on your own roles. ConnectDevs works alongside your ATS and does not replace it.

  • AI Recruiter vs Human Recruiter: Who Does What in 2026

    AI Recruiter vs Human Recruiter: Who Does What in 2026

    AI Recruiter vs Human Recruiter: Who Does What in 2026

    The AI recruiter vs human recruiter debate is usually framed as a fight, and that framing is wrong. Software now handles a large share of the repeatable work in hiring, but it does not close candidates or earn their trust. The practical question is not who wins. It is which tasks each one should own. This guide gives a clear split, a day-in-the-life example for a recruiter running 15 requisitions, the real risks, and a rollout path you can start this quarter.

    TL;DR

    • An AI recruiter is software that runs sourcing, screening, scheduling and first-round interviews. It does not replace the recruiter’s judgment or relationships.
    • AI should own repeatable, high-volume work. Humans should own persuasion, calibration, closing and anything a candidate will remember for years.
    • Candidate trust is the weak spot: Greenhouse’s 2026 survey found 38% of candidates abandoned a process because of an AI interview.
    • Regulation is moving but not disappearing. Plan for disclosure, human review and audit trails now.
    • Start with one workflow, measure it, then expand.

    Table of Contents

    What is an AI recruiter?

    An AI recruiter is software that performs recruiting tasks on its own or with light supervision: finding candidates, screening them, scheduling, and in some tools running a first-round interview. It works from a job brief and returns ranked candidates, transcripts or reports for a person to review.

    That makes it different from an applicant tracking system, which stores records. It is also different from a chatbot that answers FAQs. An AI recruiter takes an action, then hands the result to a human at a defined checkpoint.

    If you want the product-side view of what these tools automate, read our breakdown of what AI recruiting agents actually automate and what still needs a human. This article stays on the people question: what happens to the recruiter’s job.

    Will AI replace recruiters?

    No, but it will replace a lot of what recruiters spend their week doing. Those are different things, and the gap between them is where careers get made or lost.

    The pressure on teams is real. Greenhouse’s 2026 benchmark, discussed in RecTech Media, reports applications per job rising from about 115 in 2022 to 244 in 2025. Over the same period, recruiting team headcount fell 55%, while monthly hires per recruiter rose 122%. Fewer people are doing more, and tooling is part of how.

    My position is simple. Recruiters who spend their time on résumé triage and calendar tennis are exposed. Recruiters who spend it on intake, assessment, selling and closing are not.

    LinkedIn’s Future of Recruiting report points the same way. It says recruiters using AI save roughly 20% of their work week, and that demand for “relationship development” as a recruiter skill grew 54x year over year in paid job postings. When the mechanical work gets cheaper, the human work gets more valuable.

    AI recruiter vs human recruiter: who does what

    Use this as a starting split. The rule behind it: give AI the work that is repeatable, high-volume and easy to check. Keep humans on work that depends on context, trust or consequences.

    Task AI recruiter Human recruiter
    Intake and role calibration Drafts a search from the job description Owns it. Pushes back on unrealistic briefs, defines what “good” means
    Sourcing Searches large profile pools, ranks matches, enriches contacts Reviews the shortlist, spots non-obvious fits, decides who is worth a personal note
    Outreach Drafts and sequences first touches, follows up Writes to senior or hard-to-reach candidates, handles replies with nuance
    Screening Applies consistent criteria to large applicant volumes Reviews edge cases and rejections that could be wrong
    Scheduling Fully automates availability, invites, reminders Steps in only when something breaks
    First-round interview Runs structured, consistent screens and produces transcripts and scores Reads the output, decides who advances
    Later-round interviews and assessment Supports with notes and summaries Owns it, with hiring managers
    Selling the role Little to offer Owns it. Motivation, concerns, career story
    Offer, negotiation and closing Little to offer Owns it
    Hiring manager relationship Reports and reminders Owns it. Trust, pushback, expectation setting
    Compliance and final decisions Logs and flags Accountable for every decision

    Two things stand out. First, AI covers the top of the funnel almost entirely, which is where most hours go. Second, humans keep every step where a candidate decides whether to trust you.

    Where an AI recruiter wins

    • Speed and coverage. It can search and rank thousands of profiles while you sleep.
    • Consistency. It asks every candidate the same structured questions and records the answers.
    • Availability. Candidates can interview at 9 p.m. in their own time zone.
    • Admin. It does not forget the follow-up.

    Where a human recruiter wins

    • Persuasion. A passive senior engineer does not move because of a well-formatted email.
    • Judgment under ambiguity. A career gap, a pivot or a strange resume often needs a conversation, not a score.
    • Relationships. Hiring managers and candidates both remember who treated them well.
    • Accountability. A regulator or a rejected candidate wants a person who can explain the decision.

    A day with 15 requisitions

    This is an illustrative example, not customer data. Picture a recruiter in a 200-person company with 15 open roles: six engineering, four sales, three operations and two design.

    Before: the manual version

    The morning goes to inbox triage and scheduling across a dozen candidates. Midday disappears into résumé review for the two roles with the biggest applicant piles. The afternoon is sourcing for one hard role, usually the loudest one. Rejections, notes and ATS updates fill the evening.

    The result is predictable. Every req gets a sliver of attention, and the roles that need real selling get the least.

    After: AI handles the repeatable work

    • 8:30. Overnight, the AI recruiter ran sourcing for all 15 roles and completed first-round interviews with candidates who accepted invites. The recruiter reads ranked shortlists and interview reports, not raw applications.
    • 9:30. A 30-minute intake call with a hiring manager who wants to change the brief on a design role. This is human work. The AI search is updated afterward.
    • 10:30. Review of borderline candidates the system flagged. The recruiter overrides two rejections and advances one non-traditional profile.
    • 11:30. Personal outreach to three senior engineering candidates. These messages are written by hand, with a specific reason to talk.
    • 1:00. Two live interviews with finalists, then calibration debriefs with managers.
    • 3:00. A closing call with a candidate holding a competing offer. The recruiter listens for what is really driving the decision.
    • 4:30. Ten minutes checking pipeline health: which reqs are stalling, which sources convert.

    The recruiter did not become unnecessary. The recruiter’s calendar moved from processing to persuading. If you want numbers on that shift, see our piece on recruiter productivity and placements per recruiter.

    Risks: bias, compliance and candidate trust

    This is where a lot of AI recruiting content goes soft. Do not skip it. The risks are why the human stays in the loop.

    Candidate trust

    Candidates are noticing AI, and many do not like how it is used. Greenhouse’s May 2026 survey of 2,950 candidates found:

    • 63% had experienced an AI interview.
    • 38% abandoned a hiring process because of one, and another 12% said they would.
    • 70% were not clearly told upfront that AI would evaluate them.
    • 51% received no feedback afterward.

    Read that last set carefully. Much of the damage comes from surprise and silence, not from the technology itself. Tell candidates what is happening, explain what happens next, and give feedback. Those fixes cost almost nothing.

    Bias

    AI does not remove bias automatically. It can apply a flawed criterion consistently, at scale. Structured, job-related questions and scoring help, because they limit what the system can weigh. They do not replace testing your outcomes by group and reviewing rejections.

    The same Greenhouse survey found that 36% of candidates perceived age bias from both AI and human interviewers. Humans are not a bias-free baseline either. The goal is a process you can inspect and fix.

    Compliance

    The rules are shifting, so check the status before you rely on any of this. Here is where things stood as of the sources below.

    • New York City Local Law 144 requires bias audits and notices for automated employment decision tools, with penalties of up to $1,500 per violation per day. A December 2025 state comptroller audit called enforcement “ineffective,” and the city agreed to improve it. DLA Piper summarizes the audit. Weak enforcement today is not a reason to ignore the law.
    • EU AI Act. Hiring tools are classed as high-risk. Under the Digital Omnibus agreement, obligations for stand-alone high-risk systems moved from August 2, 2026 to December 2, 2027. Ogletree reports the Council gave final approval on June 29, 2026.
    • Colorado. The original AI Act will not take effect June 30, 2026. Colorado enacted SB 26-189 on May 14, 2026, a narrower automated decision-making law effective January 1, 2027. Per Seyfarth, employers must give pre-use notice and post-adverse-outcome explanations, and offer a request for human review.

    The pattern across all three is the same: tell people AI is involved, keep records, and keep a human able to review decisions. Build for that and you are covered in most places. This is general information, not legal advice, so confirm with counsel for your jurisdictions.

    A practical adoption path

    Do not roll out an AI recruiter across every role on day one. Use stages.

    1. Pick one painful workflow. Scheduling and first-round screening usually pay back fastest, especially for high-volume roles. Our guide to screening high-volume applicants covers the mechanics.
    2. Write the human checkpoints first. Decide who reviews what, and which decisions a machine never makes alone. Rejections are the usual one.
    3. Disclose to candidates. Say when AI is used, what it evaluates and how to reach a person.
    4. Pilot on two or three roles. Track time to first interview, candidate drop-off, pass-through rates and recruiter hours saved.
    5. Audit outcomes. Compare pass rates across groups and sample rejected candidates monthly.
    6. Expand sourcing next. Once screening is stable, add AI-assisted sourcing and outreach, keeping personal messages for senior hires.
    7. Redeploy the saved time. If the hours do not move into intake, selling and closing, you captured the cost saving but not the value.

    For sourcing, ConnectDevs Scout turns a role brief into a structured search and returns ranked candidates with explanations for the recruiter to review. For first-round screens, the AI Interviewer runs voice interviews and generates transcripts and structured reports. If you are comparing options, our list of best AI interview software is a good place to start.

    FAQ

    What is an AI recruiter?

    An AI recruiter is software that handles recruiting tasks such as sourcing, screening, scheduling and first-round interviews. It produces ranked candidates and reports that a human reviews. It supports the recruiter rather than replacing the role.

    Will AI replace recruiters?

    Not the role, but it will replace much of the repetitive work inside it. Recruiters who focus on intake, assessment, selling and closing become more valuable. Those who only process résumés and schedules are most exposed.

    What can a human recruiter do that an AI recruiter cannot?

    Humans persuade passive candidates, read ambiguous situations, build trust with hiring managers and negotiate offers. They are also accountable for decisions. These tasks depend on context and relationships, which software handles poorly.

    Do candidates accept AI interviews?

    Many do not like them when handled badly. In Greenhouse’s May 2026 survey, 38% of candidates abandoned a process because of an AI interview, and 70% were not told upfront. Clear disclosure, feedback and a human option reduce the damage.

    Is using an AI recruiter legal?

    Generally yes, with conditions. New York City requires bias audits and notices, the EU AI Act treats hiring AI as high-risk with obligations delayed to December 2027, and Colorado’s replacement law takes effect January 1, 2027. Check your jurisdictions and keep humans in the review loop.

    How should I start using an AI recruiter?

    Start with one workflow, usually scheduling or first-round screening on a high-volume role. Define human checkpoints, tell candidates AI is involved, and track drop-off and pass-through rates. Expand only after the pilot holds up.

    Can an AI recruiter make hiring decisions?

    It should not make final decisions alone. Use it to rank, summarize and flag, then have a person decide, especially on rejections. Several laws now require notice and some form of human review.

    Conclusion

    The AI recruiter vs human recruiter question has a practical answer. Let software do the repeatable work, and keep people on judgment, persuasion and closing. Teams that get this split right will hire faster without losing candidate trust.

    Be careful with the trust part. Disclose, give feedback and keep humans accountable for decisions. Then redeploy the time you save into the conversations that win candidates.

    Want to see how this works in practice? Start free with ConnectDevs and pilot AI sourcing and first-round interviews on one role.