How Recruiters Use AI: Sourcing, Screening and the Rules That Follow

Recruiters use AI in two very different ways, and the difference decides which rules apply. The first is as a drafting assistant: writing search strings, job ads, outreach messages and interview questions, with the recruiter making every decision. The second is as a decision tool that scores, ranks or screens out candidates. The first mostly raises confidentiality questions. The second can trigger bias audits, notice duties and discrimination liability.

This explainer walks through the common drafting uses with prompts, then sets out where the legal lines sit as of November 9, 2026.

The drafting uses

Building sourcing searches

Boolean strings for LinkedIn, job boards and search engines are tedious to build and easy to get wrong. AI is good at expanding job titles and skills into synonyms.

I'm sourcing for a [JOB TITLE] in [LOCATION OR REMOTE].
Must-have skills: [LIST]. Nice-to-have: [LIST]. Exclude: [e.g. agency recruiters, interns].

Write:
1. A Boolean search string for a general job board.
2. A variant for a site search engine query (site: operator).
3. A list of alternative job titles people in this role use.
Explain any operator that might not work on every platform.

The output still needs testing on the actual platform, since each one handles operators differently.

Drafting the job ad

A job ad is the first screening tool, and vague or inflated requirements narrow the pool before any software gets involved. AI is useful for separating real requirements from habits.

Here is a hiring manager's draft job description:
[PASTE]

1. Separate the requirements into "needed on day one" and "can be learned in the first six months".
2. Flag requirements that may not be job-related, such as degree requirements for work that doesn't need one, or physical requirements the role doesn't involve.
3. Flag wording that describes a type of person rather than the work (for example age-coded phrases).
4. Rewrite the ad in under 400 words with the pay range placeholder [PAY RANGE].

The placeholder is there so the recruiter checks local pay-transparency rules before the ad goes live.

Personalizing outreach

Outreach works better when it mentions something specific. Recruiters can summarize a candidate’s public profile in their own words and have the tool draft around it.

Write a short first-touch message (under 90 words) to a [ROLE] candidate.
What I noticed about their background (my summary): [2 TO 3 SPECIFIC POINTS]
The role: [ONE-LINE PITCH, LOCATION, PAY RANGE]
Tone: direct, no flattery, one clear question at the end.

Structured interview kits

Structured interviews ask every candidate the same job-related questions and score them against the same anchors. AI speeds up building them.

Create a structured interview kit for [ROLE].
Competencies to assess: [LIST 4 TO 6]
For each competency: two behavioral questions, one follow-up probe, and a 1-to-4 rating scale with a short description of what each score looks like.
Questions must relate to the job. Do not include questions about age, family plans, health, religion, national origin or other protected characteristics.

Intake and debrief preparation

Before an intake meeting with a hiring manager, a tool can turn a rough brief into a question list. After interviews, it can tidy each interviewer’s typed feedback into a common format before the debrief. The hiring decision is made by people in the room.

When a tool becomes a decision tool

UseWho decidesMain legal exposure
Drafting searches, ads, messages, questionsRecruiterConfidentiality; discriminatory wording in ads
Summarizing interviewers’ notesHiring teamAccuracy; consistency
Scoring or ranking applicantsSoftware substantially influencesBias audit and notice rules; disparate impact
Automatic rejection rulesSoftwareDisparate treatment and impact; ADEA, Title VII, ADA

Automatic rejection rules don’t need to be sophisticated to cause trouble. In September 2023 the EEOC announced that iTutorGroup would pay $365,000 to settle a suit alleging its application software automatically rejected female applicants aged 55 or older and male applicants aged 60 or older. That was a simple age filter, not machine learning.

Federal rules

The federal anti-discrimination laws (Title VII, the ADA and the ADEA) apply to hiring decisions however they are made. The EEOC took its AI-specific technical assistance offline in January 2025, and its former AI page (eeoc.gov/ai) returned a “page not found” message when checked on November 9, 2026. Removing guidance doesn’t change the statutes, and the iTutorGroup press release remains on the EEOC site.

New York City Local Law 144

New York City’s rules on automated employment decision tools (AEDTs) have been enforced since July 5, 2023. According to the city’s Department of Consumer and Worker Protection FAQ:

  • An employer or employment agency can’t use an AEDT unless it has had a bias audit within the past year.
  • The employer, not the vendor, is responsible for making sure the audit was done, though a vendor can arrange one.
  • The auditor must be independent of both the employer and the vendor.
  • A summary of the latest audit results and the date the tool was first used must be published, for example on the careers page.
  • Candidates and employees who live in New York City must get notice at least 10 business days before the tool is used. The notice must say what the tool assesses and how to request an accommodation.

State laws

  • Illinois. Amendments to the Illinois Human Rights Act took effect January 1, 2026. They bar employers from using AI that has a discriminatory effect in covered decisions or using zip codes as a proxy for protected classes, and they require notice. The Illinois Human Rights Commission said in April 2026 that the law covers employers with one or more Illinois employees for 20 or more weeks a year. Detailed notice rules were still at the proposed stage at that point.
  • California. The Civil Rights Council’s regulations on automated-decision systems took effect October 1, 2025. They clarify how existing anti-discrimination law applies when employers use AI and algorithms in employment decisions.

Other states have passed or proposed their own laws, and effective dates have moved, so recruiters hiring across states should check each one with counsel.

Don’t paste this in

Candidates’ resumes, contact details, assessment results or interview feedback into a personal AI account, and anything revealing age, disability, medical information, family plans, religion or other protected characteristics into any tool. Using software to screen or rank candidates in New York City needs a bias audit and advance notice under Local Law 144. Federal laws enforced by the EEOC apply to every hiring decision, whatever tool shaped it.

Mapping one hiring pipeline

A practical way to see where the rules bite is to walk one requisition through its stages and mark each AI touchpoint. Take a customer-support hiring round at a company with staff in several states, including some candidates in New York City.

  1. Job ad drafted with AI. Drafting help only. The recruiter checks the wording and adds the pay range. No audit duty.
  2. Applicant tracking system ranks applicants by match score. This is the stage to examine. If the score substantially assists the decision on who moves forward, it may count as an AEDT under the city’s rules, which would require a bias audit and notice to NYC residents. Illinois and California rules may also apply, depending on where candidates live and work.
  3. Recruiter phone screens with a structured kit. The AI drafted the questions; people ask them and score answers. The legal exposure is the usual one: questions must relate to the job.
  4. Interview feedback tidied by AI. Interviewers type their own notes, and the tool formats them. A person still reads every original note before the debrief.
  5. Offer. No AI decision involved.

In this pipeline, step 2 is the only point where a vendor’s audit, the notice timing and published results come into play. Ranking and matching features can come bundled with a tracking system rather than bought separately, so the inventory in the checklist below can turn up tools nobody chose deliberately.

Questions to put to a vendor

  • Does the tool score, rank, filter or recommend candidates, and can those features be switched off?
  • Has an independent bias audit been done in the past year, and can the summary be shared?
  • What data does the model use, and does it use anything that could stand in for a protected characteristic, such as zip code or graduation year?
  • Is candidate data used to train the vendor’s models, and for how long is it kept?
  • Can the employer export the information needed to give candidates notice and handle accommodation requests?

A recruiter’s checklist

  1. Keep a list of every AI tool in the hiring process, and mark which ones only draft and which ones score, rank or filter.
  2. For anything that scores or filters, ask the vendor for its bias audit, data sources and the factors used.
  3. Check whether any candidates live in New York City or apply to roles there, and whether Illinois, California or other state rules apply.
  4. Keep human review on every rejection that software recommends.
  5. Store candidate data only in approved systems.

Recruiters comparing sourcing, screening and scheduling software can see which vendors publish bias-audit information in the best AI for recruiters.



Previous Post
Best AI for Recruiters: Sourcing, Screening and Scheduling Tools Compared
Next Post
AI for HR Professionals: Seven Workflows and the Employee Data That Stays Out