Recruitment Automation: What to Automate, Which Laws Apply

Recruitment automation is software that does the repeated steps of hiring for you: sending job posts, reading applications, booking interviews, sending updates, writing interview notes and building reports. It works by rules ("when a candidate applies, send a confirmation") or by AI ("read each CV against the role's criteria"). Automate the high-volume steps that follow a rule, and keep the decision on each candidate with a person. When AI evaluates applicants, it also brings legal duties: a bias audit in New York City, record keeping in California, and notice and human-review rights in the EU.

What is recruitment automation?

Recruitment automation is the use of software to run hiring tasks that a recruiter would otherwise do by hand, step by step, for each applicant. The applicant tracking system (ATS) holds the record: who applied, at which stage they are, what was said. Automation is the work done on that record, and it comes in two kinds:

• Rules. A trigger starts a fixed action: a new application gets a confirmation email, an interview gets a reminder 24 hours before.
• AI. Software reads, evaluates or writes: it compares a CV with the role's criteria, runs a first interview or summarizes a call.

If your largest manual task is the first screen, Firstview's AI resume screening screens every CV on your criteria, with a reason for each criterion. "Nobody is rejected automatically": a recruiter decides who moves on.

Which recruitment tasks can you automate?

Recruitment automation can reach 8 hiring steps. For each one, the table splits the task the software does from the decision a person keeps. The last column names the rules to check when AI does the step; the law table below gives their dates.

Step	What software can do	What a person keeps	Rules to check when AI does it
Job posting and sourcing	Post to boards, search your own talent pool	The criteria, the decision to contact someone	EU AI Act high-risk rules, from 2 December 2027
Resume screening	Read every CV against the criteria, mark missed requirements	Who moves on and who is rejected	NYC Local Law 144, GDPR Art. 22, EU AI Act (2027)
Scheduling	Offer open slots, send reminders, rebook	Exceptions, panel make-up	No rule in the law table below
First interview	Ask the same questions, follow up, write a report with evidence	The decision after the report	EU AI Act Art. 50 notice, NYC Local Law 144
Candidate messages	Confirmations, status updates, reminders	Rejection wording, offers	No rule in the law table below
Interview notes	Record, transcribe and summarize calls into a template	The scorecard and the hiring decision	Local recording notice rules (not in the table)
Offers and onboarding	Fill offer letters from templates, route approvals, send onboarding tasks	Offer terms and the conversation with the candidate	No rule in the law table below
Reports	Pass-through rates, time per stage, drop-off	What to change in the process	California: keep automated-decision data 4 years

How much time does recruitment automation save?

Recruitment automation saves the most time at the screening step, because that is where the volume is. Greenhouse's recruiting benchmarks, based on more than 6,000 companies and 640 million applications, count 244 applications per job in 2025, up from 116 in 2022. If a recruiter spends 2 minutes on each CV (our assumption, not a measured number), one job takes about 8 hours of reading before the first call.

Scheduling saves less time than screening. In Ashby's 2026 recruiting operations report (54 million applications and 93,000 jobs, January 2021 to March 2026), automated scheduling took a median of 3.7 hours to confirm an interview, against 5 hours by hand. That is about 1.3 hours per interview: a real gain, but small next to Ashby's median time to first fill of 60 days for business roles.

Automated updates also help candidates. In Greenhouse's 2026 Candidate AI Interview Report (2,950 job seekers in the US, UK, Ireland, Germany and Australia), 51% of US candidates who completed an AI interview never received an outcome. An automated status message at each stage is a direct way to tell candidates where they stand.

Automation also does not fix judgment. In the same Ashby data, about 38% of scorecard pairs differ by at least one point between interviewers. To narrow that gap, use a structured interview with one set of criteria for every interviewer.

To estimate your own hours, the screening time calculator works out monthly hours on CV review, booking and screening calls from your volumes.

Which laws apply to recruitment automation?

Laws on recruitment automation target the steps where software evaluates candidates; reminders and booking fall outside the rules below. This table lists the rules that an employer in the US or the EU meets most often, with the date each duty starts:

Jurisdiction and rule	When it applies	What it requires	Starts
New York City, Local Law 144	An automated employment decision tool used for candidates or employees in the city	The employer: a bias audit within one year of use, public information about the audit, and a notice to candidates at least 10 business days before use	5 July 2023 (enforcement)
California, FEHA regulations	Any automated-decision system used in employment	A system that harms applicants on protected characteristics may violate the law; employers keep records, including automated-decision data, for 4 years	1 October 2025
EU AI Act, Article 50	An AI system that talks directly with people, such as an AI interviewer	The provider designs the system so that candidates know they are talking with an AI, unless it is obvious	2 August 2026
EU AI Act, Annex III and Article 26	AI used to place targeted job ads, analyze and filter applications or evaluate candidates	The employer: human oversight by trained staff, use as the provider instructs, keep logs at least 6 months, inform candidates and workers' representatives	2 December 2027
EU GDPR, Article 22	A decision based only on automated processing with a legal or similarly significant effect	Such decisions are restricted; where one is allowed, the candidate keeps the right to human intervention and to contest the decision	Already applies
Colorado, SB 26-189	Automated decision-making technology used to materially influence a consequential decision	Duties for developers and deployers; the Attorney General's rules are due before the law takes effect	1 January 2027

Three points matter for planning:

1. The EU timeline moved. The AI Omnibus amendment, in force since 27 July 2026, moved the high-risk rules for employment from 2 August 2026 to 2 December 2027. The Article 50 notice already applies. Our EU AI Act guide for AI interviews has the detail.
2. A person's review must be real. GDPR Article 22 covers a decision "based solely on automated processing". If a recruiter confirms every list the software produces without reading it, the review adds little, and a regulator may see the decision as automatic.
3. Telling candidates is also good practice. In Greenhouse's 2026 Candidate AI Interview Report (2,950 job seekers in the US, UK, Ireland, Germany and Australia), 70% of US candidates who had an AI evaluation said it was not clearly disclosed before their last AI interview. When US candidates left a process, a company's failure to disclose how AI would be used was one of the top reasons (27%).

This table gives no legal advice. Check the rules of each place where you hire with your counsel before you automate a screening step.

What should stay with a person?

The decision on each candidate should stay with a person: who moves on, who is rejected and who gets the offer. Software can prepare that decision with evidence. The top-ranking guides we read from IBM, Greenhouse and Recruiterflow say the same, and the EU rules above require human oversight or human review.

Keep these steps human:

• Setting the criteria. The requirements, nice-to-haves and dealbreakers for a role come from the hiring manager. An intake meeting is where they are agreed.
• Rejections. Let software mark a missed requirement; let a recruiter confirm the rejection. GDPR Article 22 restricts a rejection made only by software, and where one is allowed, the candidate in the EU can ask for human intervention.
• Unclear cases. A CV that does not show a criterion, or an interview answer that the software cannot place, goes to a person for review.
• Accommodation requests. A candidate who asks for a different format needs a person who can change the process.
• The final hire. Interviews, references and the offer conversation stay human.

How to automate your recruitment process in 7 steps

A low-risk way to automate recruitment is to start with rule-based steps, add AI where the volume is, and measure each change against the numbers you had before.

1. Map your current process with numbers. For one role, write down each step, who does it, how long it takes and how many candidates pass. Our hiring funnel benchmarks give pass-through rates to compare with.
2. Automate the rule-based steps first. Confirmations, status updates, reminders and self-scheduling carry little risk, and none of them evaluates a candidate.
3. Check the law table and your tools before you add AI. For each place where you hire, list which duties a screening tool or an AI interview brings: an audit, a notice, record keeping, human review. Then check how results reach your ATS, calendar and email; every hand-off between tools is a place where a candidate can get lost.
4. Pilot AI screening on one high-volume role. Run the software and your recruiters side by side for 2 weeks. Compare who each one would advance and read every case where they disagree.
5. Train the team before the full launch. Show recruiters and hiring managers how to read a result and where to override it. Agree on one rule: a recruiter reads every case before a rejection.
6. Tell candidates what the software does. Put one sentence in the invitation: which step uses AI, what it evaluates and that a person makes the decision.
7. Measure 4 numbers for 4 weeks before and 4 weeks after. Time from application to first contact, pass-through rate per stage, drop-off per stage and offer acceptance. Keep a change when time to first contact falls and none of the other 3 numbers gets worse.

Sample: an automated first-round summary that a person reviews

The output of an automated first round should give a recruiter the evidence for a decision and leave the decision itself open. This is a sample for a fictional candidate, in the verdict format of Firstview's interview report (for example Exceeded, Met, Partially met or Unclear for each criterion). You can run the same check on the output of any tool you use.

Sample. Role: customer support agent, evening shift. Candidate: "Candidate A" (fictional).

Criterion	Verdict	Evidence (candidate's own words)
Can work evenings and weekends	Met	"I can do evenings from Monday to Saturday, and Sundays every other week."
Calms an upset customer	Exceeded	"She was shouting about a double charge. I said I could see both payments and stayed on until the second one was reversed."
Uses a ticketing system	Partially met	"We logged calls in a shared spreadsheet, not a real ticketing tool."
Written English for chat support	Unclear	None: a voice interview does not test writing.

Decision: to be made by the recruiter.

Check the output with 4 questions:

1. Does every verdict other than "Unclear" quote the candidate, so you can see what it rests on?
2. Does "Unclear" go to a person?
3. Is the decision field empty until a person fills it?
4. Could you explain this result to the candidate if they asked?

A "no" to question 2 or 3 means the tool is making decisions, and the rules above apply with more weight. A "no" to question 1 or 4 means a recruiter cannot check the result. The interview scorecard template shows how to turn verdicts like these into one rating per criterion.

How Firstview answers the 4-question check

Firstview runs the screening and first-interview steps, so the 4 questions above apply to it as well. These are the answers from Firstview's product pages:

Question	Firstview's answer
1. Does each verdict show its evidence?	"Firstview's AI interview report gives a verdict for each criterion with the candidate's own words behind it." CV results have "a reason for every criterion".
2. Where does "Unclear" go?	"'Unclear' and 'Needs review' go to a person, never to an automatic rejection."
3. Who makes the decision?	"No. It writes the report. A person decides." In CV screening, "A person moves candidates."
4. Can you explain the result?	Firstview "explains every result", and a missed dealbreaker only marks the CV.

The AI interview software runs live voice interviews in 42 languages, and "nobody has to schedule a call". The AI note taker for recruiters "joins your calls, records, transcribes and summarizes them, writes notes in your template and files them on the right candidate."

How we made this guide

We read 5 of the top 6 Google results for "recruitment automation" in the US (Semrush, 2026-11-04; a Reddit thread at position 5 could not be loaded) and wrote down which steps, numbers and rules each one covers. The law table is our own cut of primary sources, read on 2026-11-04: the NYC Department of Consumer and Worker Protection, the California Civil Rights Council, the EU AI Act Service Desk, the European Commission and the Colorado Attorney General. The numbers come from Greenhouse's 2025 benchmarks and 2026 candidate report and Ashby's 2026 operations report. The sample summary is fictional and uses the verdict format of Firstview's interview report. We checked each Firstview fact against Firstview's product pages on 2026-11-04.