The most automated desk in HR is the most sued
Recruiting is where HR's AI adoption concentrates — SHRM's 2025 numbers again: 51% of organizations use AI to support recruiting, more than any other HR practice area. The task breakdown reads like your Monday: writing job descriptions (66% of AI-using recruiters), screening resumes (44%), automating candidate search (32%), customizing postings (31%), applicant communication (29%). And 89% of the recruiters using it say it saves time.
Hold that 89% next to another number: Gartner's late-2025 finding that 88% of HR leaders say their organizations have not realized significant business value from AI tools. Both are real, and they don't contradict — time saved on a task and value realized by an organization are different quantities. The gap between them is exactly what this lesson closes: a JD drafted in four minutes creates value only if what it feeds — the screen, the brief, the decision — can be trusted. An unverified screening summary doesn't save time; it moves the time to whoever has to untangle it later, possibly in a deposition.
Because recruiting is also where the legal surface is sharpest. Resume screening is the 44% — and automated screening is the act that NYC audits, Illinois requires consent for, Colorado will require human review of, and Mobley v. Workday is litigating at the scale of a billion rejected applications. The stakes aren't hypothetical for small teams either: the liability theory in Mobley runs through the vendor, which means it doesn't care what city you hire in or whose logo is on the tool.
So this lesson holds one line, and builds the code that holds it for you:
AI drafts the brief. Every claim in the brief carries a receipt. The decision belongs to a named human.
A receipt is a pointer to the actual resume — the line that proves the claim. "Ran multi-state payroll" cites the line that says so. "Probably great under pressure" cites nothing, because nothing on the resume says it — so it's either labeled inference and weighed by a human, or it's out. What a screening brief may never contain is the third thing: an inference wearing the costume of a fact. That's the exact material a plaintiff's lawyer reads aloud slowly.
The next six steps make the receipt a data structure, the audit a loop, and the "no unlabeled inference" rule a gate that holds two briefs at the checkpoint. Your ATS doesn't do this. That's why you're here.