Six recruiting jobs worth handing to AI this quarter (and the one to keep human)

Recruiting is mostly admin wrapped around a few real judgements. The judgements are what clients pay you for. The admin is what eats the week. In 2026, AI is good at the admin. Good enough that recruiters using it report getting three to five hours a day back, cutting cost-per-hire by roughly a third, and filling roles about a quarter faster.

So this quarter is a good time to hand some of it over. With a catch on the last job on the list: a couple of these tasks touch a decision about a candidate, and from December those are regulated, with the liability sitting on your agency, not on the software.

The split looks like this.

Hand these over this quarter

None of the jobs below makes a decision about a person. They draft, search, and summarise. Low risk, high reward, and the place to start.

Writing job ads. A decent model writes a clean, inclusive ad in seconds, and it beats most humans at catching the loaded language that quietly stops people applying. Give it the role, the must-haves, and your tone, then edit the output. Trial this one first. We put a free version of it online for exactly that.

Building search strings. Boolean by hand is a black art most recruiters half-remember. Describe the person in plain English and let the model build and tighten the search string. Faster, and it surfaces combinations you would have missed.

Drafting outreach. Personalised messages at volume is the classic win. Feed it the candidate's background and the role, get a first draft per channel, then edit the ones worth the extra touch. Reply rates live and die on that edit, so keep your hand in.

Summarising references and interview notes. Record with consent, ask a model to summarise, and an hour of writing up becomes a two-minute review. The summary is a draft you check, not a record you trust blindly.

Building interview scorecards. Ask for a structured scorecard tied to the role's actual competencies. Structured interviews are fairer and more predictive than gut feel, and AI turns building one into a five-minute job instead of the chore everyone skips.

That is most of a recruiter's admin week, handled, with nothing that decides anything about a candidate.

The one to be careful with

Then there is CV screening. Ranking, shortlisting, auto-rejecting.

It is the biggest time-saver on the list, which is exactly why it is the one to slow down on. Screening a stack of CVs is the moment a decision gets made about a person. From 10 December 2026, that decision is a regulated one: the Privacy Act's new automated-decision rules explicitly cover recruitment, because progressing or rejecting a candidate plainly affects their interests. If software does the screening, you have to be able to say so, and say how.

Why "the tool did it" won't save you

The part the vendors skip in the demo: if an AI screening tool systematically disadvantages candidates on a protected attribute, the agency wears the claim. Not the software company.

Under the Fair Work Act and state anti-discrimination law, using a tool doesn't move the liability. The Australian Human Rights Commission has been plain that AI in hiring still has to satisfy the Sex Discrimination Act and the Racial Discrimination Act. And the risk isn't theoretical. These tools learn from your past hiring, so if your history leans a certain way, the model learns to lean the same way, faster and at scale.

"We used an approved tool" is not a defence. "We can show how the decision was made, and a person made it" is.

Doing the sensitive jobs right

Defensible screening isn't complicated. Four things. A human makes the actual call, with a name against the progress-or-reject decision. Protected attributes, and the obvious proxies for them, stay out of the screening logic. You can explain in plain words why any candidate was moved forward or not. And there's a record of who decided and on what basis.

Do that and you get the time back without the exposure. Skip it and you've automated a discrimination claim.

Where to start

Pick two or three of the admin jobs and trial them this week. Job ads and outreach drafts are the fastest to feel. Leave screening human-led until the four safeguards are in place.

If you want a quick read on where your current candidate tools sit against the December rules, we built a free AI Exposure Check. A few plain questions, and you get a map of what's caught and what to fix first. About three minutes.

Run the free AI Exposure Check →

This is general information about a moving regulatory picture, not legal advice. For a screening process you're going to rely on, get it checked for your agency.