AI is widely deployed across the recruiting funnel in 2026. Resume screening tools are in the majority of enterprise ATS systems. AI job description writers are standard features in LinkedIn Recruiter and Greenhouse. Interview scheduling chatbots handle the back-and-forth that consumed hours of recruiter time. The tools are useful. The risks are real. Knowing the difference between the two is what separates effective AI-assisted recruiting from the systems that expose companies to legal liability and miss the best candidates.
Current Applications of AI in Recruiting
Resume screening and ranking. The highest-adoption AI application in recruiting. AI scoring tools rank candidates by likelihood of success using features extracted from resumes: skills mentioned, companies worked at, tenure length, education credentials, gaps between roles. High adoption does not mean high reliability -- this is also where the most significant AI bias problems have been documented.
Job description writing. AI writing tools generate first drafts of job descriptions from a brief: role title, key responsibilities, required skills, seniority level. This is one of the cleaner AI recruiting applications. The risk is low (a job description is reviewed before posting), and the time savings are real. The main downside is that AI-generated JDs tend to be generic -- if every company uses AI to write JDs, the output becomes homogeneous.
Interview scheduling. AI scheduling chatbots handle the logistics of finding a time that works for all parties, sending calendar invites, and sending reminders. This eliminates a significant time drain for recruiters. The automation is reliable and the failure modes are low-stakes (a scheduling error is easily corrected).
Interview question generation. AI generates structured interview question sets from a job description and role competencies. Useful for standardizing interviews across different interviewers -- gives less experienced interviewers a concrete guide rather than leaving them to improvise. Quality varies; the best output comes when you give the AI specific competencies to assess rather than asking for "good interview questions."
Candidate sourcing. AI tools scan LinkedIn profiles, GitHub accounts, and professional databases to identify candidates who match a role's requirements without having applied. This expands searches beyond the active candidate pool and surfaces people who would not have found the posting. High value for technical roles where the best candidates are often passively employed.
The Bias Problem
Amazon's AI hiring tool is the canonical failure case and is worth understanding in detail. Amazon built a machine learning model to rank engineering candidates. The model was trained on historical hiring decisions -- which candidates Amazon had hired over the preceding ten years. Those historical decisions reflected a workforce that was heavily male. The model learned that being male was correlated with being hired. It penalized resumes that contained the word "women" (as in women's coding club, women's chess team) and downgraded graduates of all-women's colleges.
Amazon scrapped the tool in 2018. But the problem was not specific to Amazon's implementation -- it is structural. Any AI model trained on historical hiring decisions will learn whatever patterns were in those decisions, including patterns that reflect bias. This is not a solvable problem through better engineering alone. It requires deliberate bias testing before deployment.
The EEOC (Equal Employment Opportunity Commission) and multiple state governments have issued guidance that AI hiring tools must be validated for adverse impact. This means testing whether the tool produces different selection rates across protected groups (race, gender, age, disability status). If it does, the tool must be justified under business necessity standards or modified to eliminate the disparity.
This is not a theoretical concern. Illinois and New York City have enacted AI hiring laws with enforcement mechanisms. Federal rulemaking is ongoing. Companies using AI screening tools without adverse impact testing are carrying legal risk.
