A note before the split. This is general guidance rather than legal or employment advice. Candidate data protection rules, discrimination law and any regulations on automated decision-making in your jurisdiction override everything below, as does your organisation's policy. Tools change constantly, and so, currently, does the law around AI in hiring.
Recruiting has become the first profession where both sides of every interaction arrive AI-assisted, and it happened without a memo. The cover letter that reads suspiciously well probably is, the CV has been polished by the same tools you use, and the candidate rehearsed your likely interview questions against a chatbot on the train.
None of this is cheating, any more than your AI-drafted outreach is, but it has quietly broken something recruiters relied on for decades: polish used to be a signal, and now it is a default. A well-written application no longer tells you the candidate writes well. It tells you they have an internet connection.
That reshapes both halves of the job, what you delegate and what you now have to detect, so this guide covers both.
The short answer: trust ChatGPT with the production side of recruiting, including job ad first drafts, search strings, outreach skeletons that you then make genuinely specific, interview question sets, and the candidate-communication volume that fills the gaps between conversations. Hold two lines that carry legal weight rather than just professional weight: no identifiable candidate data on consumer versions, and no hiring decision, screen-out or rejection reasoning generated by the tool, because bias and accountability in hiring are regulated matters and the human decides. Then recalibrate what signals mean, since substance now has to do the work polish used to.
What to hand over
| The job | What that looks like |
|---|---|
| Job ad first drafts | The role turned into a readable, honest ad in minutes, with your edit for accuracy and your check that nothing in it skews who feels invited to apply |
| Search craft | Boolean strings, keyword variants and synonym sets for a role, one of the most quietly effective uses in the profession |
| Outreach skeletons | The structure and boilerplate handled, with the one genuinely specific line about this candidate written by you, since candidates smell the merge field too. The same edit that stops AI emails reading as AI applies here. |
| Interview design | Question sets probing for the competencies the role actually needs, structured consistently so candidates are comparable |
| The communication volume | Scheduling, updates, the gentle chase, and the humane version of hard news, drafted for your review and your name |
The two lines with legal weight
Most professions get a data rule. Recruiting gets a data rule and a decision rule, and both carry more than reputational stakes.
The data rule comes first. Applications, CVs and interview notes are personal data about identifiable people who did not consent to feeding a consumer chatbot, and everyday versions of these tools can learn from what goes into them. Candidate material lives in your ATS and sanctioned systems, and when you want AI help thinking about a candidate situation, it travels as an anonymous shape, a strong applicant with an unexplained gap, never as a named person. The hypothetical method is the practical version of that rule.
The decision rule matters even more. The tool will cheerfully rank CVs, recommend rejections and generate reasons, and letting it is a problem stacked three deep: its judgements can encode bias you cannot see, several jurisdictions now regulate automated hiring decisions specifically, and a rejection reason you did not actually reason yourself is indefensible the day anyone asks.
AI can prepare information for a hiring decision. A person makes it, and can say why. The same distinction applies across the people function, which is why the HR version of this question places its hard line in much the same territory.
Reading candidates in the AI era
The detection half deserves its own honesty, because the instinct to hunt for AI-written applications is mostly a trap. Detection is unreliable, punishing candidates for using the same tools you use is hard to defend, and polish was never the thing you were actually hiring for anyway.
The productive response is to move your signal-reading to what AI assistance cannot fake: the specific example under follow-up questioning, the reasoning a candidate can walk through out loud, and the details of what they actually did that survive a "tell me more".
Interviews shift weight accordingly, from reviewing what was written to probing whether the person can stand behind it. It is the same standard this whole series applies to AI output in every profession. The candidates worth hiring can defend their material, and so can good recruiters.
The line that holds
The arrangement that works keeps the two halves of the job in their lanes. The tool produces at volume, the ads, the strings, the skeletons, the questions and the updates, then hands recruiting hours back to the part that was always the craft: judging substance, reading people, and making decisions you can stand behind and explain.
The profession's paradox is that AI made everyone's materials better and everyone's signals worse. The recruiters doing well inside it are the ones who stopped grading the polish and started probing what is underneath, with the machine handling everything that never needed a human in the first place.
Not sure where to start?
Volume from the tool. The judgement calls stay yours.
The free finder tells you which AI to open for the task in front of you, in about 60 seconds, and the full "AI, sorted." reference lands in your inbox.
Try the free finder →One question in, one answer out.
Clair helps non-technical professionals know when to trust their AI, when to check it, and when to skip it.