AI for Recruiters in 2026: Intake to Offer, and the Hiring Laws

AI can handle much of a recruiter's paperwork today: draft Boolean strings, write outreach, turn screening calls into scorecard notes, juggle interview panels and tidy the ATS. It should not decide who advances. Tools that sort or screen applicants also carry legal duties in New York City and Illinois, soon Colorado, and later the EU.

This guide follows one requisition from the intake meeting to the offer letter, in-house and agency, then sets out where each hiring law stands in November 2026. Those laws have moved a lot in the past year, so a summary from last spring may already be out of date.

Where a recruiter's hours go

A recruiter's hours mostly go to moving information between people and systems. The BLS description of the job covers finding applicants, screening and interviewing them, contacting references and extending offers. O*NET's task list adds the less visible parts, such as keeping applicants updated on where they stand and coordinating with outside agencies.

On a live req, that looks like this: a hiring manager says "someone like the last hire, but more senior," and you turn it into a search. You run the strings, send the messages, chase replies, take the screens, write them up, and find an hour when four interviewers and a candidate are all free. Then you collect late scorecards, get the offer approved, and tell the others no. Every step leaves a record that has to land in the applicant tracking system.

Recruiting tasks AI can take, one by one

AI is most dependable on tasks that produce a draft you can check, and riskiest on tasks that sort people. Prices below are as published by each vendor; several sell only by quote.

Boolean is a good first test because errors are visible. LinkedIn's own community examples use AND, OR and NOT in capitals, with parentheses and quoted phrases. Ask for that syntax, then run the string and read the first page of profiles. If the results are wrong, fix the brief before the string.

Here are my notes from an intake call for [role]:
[paste notes]

1. List must-haves, nice-to-haves and anything the hiring manager contradicted.
2. List job titles this person may hold at other companies.
3. Write one LinkedIn Recruiter Boolean string using AND, OR, NOT, parentheses and quoted phrases.
4. List questions I should take back to the hiring manager.

Use only skills and experience. Do not use names, schools as a proxy, graduation years, ZIP codes or anything that hints at age, sex, race or disability.

In-house and agency recruiters need different tools

In-house recruiters usually start with the AI already inside their ATS, because the ATS is the system of record and hiring managers already work there. Check what your Greenhouse or Ashby contract includes before buying anything. LinkedIn's Hiring Assistant, an extra on top of LinkedIn Recruiter with no listed price, covers sourcing, evaluating applicants, outreach and pre-screens.

Agency recruiters live on speed to submittal and on client relationships, so staffing-built tools fit better. Bullhorn Amplify runs background screening and surfaces ranked matches from your own database. Loxo lists a Candidate Submittal agent; plans start at $149 a user a month on annual billing, with the agents on its quote-only Professional plan. Recruit CRM lists candidate submission and pitching agents from $95 a user a month, and resume formatting at $135.

Being the agency doesn't move you outside the rules. New York City's FAQ names employment agencies alongside employers. Before you run a client's req through a screening agent, settle in writing who handles the audit and the notices.

AI hiring laws for recruiters as of November 2026

US AI hiring rules now come mainly from cities and states, while federal enforcement has pulled back without changing the statute. Treat this as a map for a talk with employment counsel, not legal advice. Your duties usually follow where the job or the candidate is, not where your desk is.

Three more points belong on your list. Illinois has a separate AI Video Interview Act for AI analysis of recorded interviews. California's rules on automated-decision systems in employment took effect October 1, 2025, and require four years of records. And vendors can be pulled in too: in Mobley v. Workday, a federal court in May 2025 preliminarily certified an age discrimination collective over AI screening. That order is preliminary and the claims are unproven.

What stays with the recruiter

Judgment about people stays with the recruiter, and the vendors say so themselves. Loxo's line is "Never through the decision." LinkedIn says you "remain in control." Those promises only hold if your process does.

• Pushing back at intake. A model will write up an impossible req neatly. Only you can tell the hiring manager that the salary and the seniority don't match.
• Reading a candidate. Hesitation about a relocation, or a counteroffer they haven't mentioned, comes out in conversation.
• Closing and declining. The offer call and the no to a finalist who met the whole team are yours to make.
• Recording consent. Federal law asks for one party's consent, but about 11 states, including California, Illinois and Florida, primarily require everyone's. Say you're recording and get agreement at the top of every screen.

Chatbots aren't a way around any of this. If you paste 40 resumes into one and ask for the top five, the same notice and audit questions may follow, because they turn on what the tool does in your decision.

How Merlin helps on a recruiting desk

Merlin, which runs on a Mac, handles the conversations around a req and isn't a recruiting system. It runs on Gmail and Google Calendar accounts, so it won't suit an Outlook-only desk. It doesn't source, rank or screen candidates, and it doesn't phone or text them. It has no built-in ATS connection, though you can add a custom MCP link (a common link between outside software and AI tools) if your system offers one, as Loxo and Recruit CRM do.

Merlin's use here is follow-through. The People panel flags candidates and hiring managers owed a reply, read from your email, so a message from last week doesn't get lost. It also shows contacts who went silent after regular emails, which suits a warm pipeline. Shortly before a meeting, Meeting Brief pulls together attendee backgrounds and prior discussions.

For screens held on your Mac, Merlin records the call on every plan, and transcripts naming each speaker need a paid plan, from $29 a month. An Interview template turns them into structured notes, and you paste what belongs into the ATS yourself. The recording is saved locally, but paid transcription uploads the audio, so get the candidate's consent and check your company's policy first. A note to a candidate requested in chat arrives as an editable draft that waits for your confirmation. Routines, Merlin's scheduled tasks, run unattended, so keep them for your own digests rather than candidate messages.

How to try this on one req this week

1. Sort your stack into two lists. One holds features that draft text for you. The other holds anything that scores, tiers, ranks or screens people, including features already switched on in your ATS.
2. Map jobs and candidates to rules. Note any open role or candidate in New York City, Illinois, Colorado, California or the EU, and take the second list to counsel with that map.
3. Run one live req through the prompt above. Compare its string with your own and see which one finds people you'd contact.
4. Try a notetaker on your next five screens, with a consent line at the start. Check each summary against your memory of the call before it reaches a scorecard. If you're comparing options, see these AI note takers.
5. Ask each screening vendor four things: the date of its latest bias audit, who ran it, whether candidates can opt out, and what it logs about each decision.