AI for Hiring and HR — Learn AI

By the end of this lesson you will be able to

Write a job advert that describes the actual job
Build a structured interview instead of improvising
Know exactly where the line is on assessing candidates, and why

The line, stated plainly

Do not use AI to rank, score, or reject candidates, and do not use it to write assessments of named employees without a manager's input and review. Beyond the legal exposure in several jurisdictions, models reproduce patterns from their training data — and hiring is exactly the domain where those patterns encode historical discrimination.

That rule removes the use everyone asks about first and leaves a surprising amount of genuinely valuable work — most of which is about being more consistent and more explicit, which happens to be what makes hiring fairer anyway.

Job adverts that describe the job

Most job adverts are assembled from other job adverts, which is why they all list "excellent communication skills" and none of them tell you what the work is actually like. Start from the work instead.

Prompt

Write a job advert for a [role] at a [size] [industry] business
in [location].

Here is what the person will actually do, week to week:

[DESCRIBE THE REAL WORK — messy is fine]

Here is what is genuinely hard about the job:

[BE HONEST]

Format: short. Role summary, a typical week, what we need, what we
offer, how to apply.

Rules:
- No "fast-paced dynamic environment", no "wear many hats",
  no "rockstar", no "we're like a family"
- Requirements must be things that are actually required. Flag any
  I listed that are probably preferences, and ask me
- Include the salary range: [range]
- Describe the hard parts honestly — we would rather filter now

The requirements check

Asking it to challenge your requirements is the most useful line. Inflated requirements are the cheapest and most common way to shrink a candidate pool — and they disproportionately deter exactly the candidates who would have been fine.

Structured interviews

Unstructured interviews are close to useless as predictors and are where most bias enters a hiring process. Structure fixes both, and AI makes structure cheap enough that small businesses can actually do it.

Derive questions from the actual work
Paste the job description and ask for questions that test the specific things the role requires, not general character questions. "Tell me about yourself" tests nothing.
Ask for a scoring rubric
What does a strong, adequate, and weak answer to each question look like? Deciding this before you meet anyone is the entire mechanism by which structured interviews reduce bias.
Use the same questions for every candidate
Non-negotiable. Comparability is the point, and it is the first thing that gets abandoned when someone is running late.
Score against the rubric yourself
A person scores. Every time. AI helped you prepare; it does not get to evaluate.

Prompt

Here is the job description for [role]:

[PASTE]

Build a structured interview:

1. Six questions that test the specific competencies this job needs.
   Behavioural where possible ("tell me about a time...").
2. For each, what a strong / adequate / weak answer contains.
3. Two follow-up probes per question for when an answer is vague.
4. Flag any question that risks straying into protected
   characteristics, and suggest a safer version.

Do not include: "where do you see yourself in five years",
"what is your greatest weakness", or brain-teasers.

Onboarding

Almost every small business onboards badly — not from indifference but because writing a 30/60/90 plan takes half a day nobody has when someone is starting Monday.

Prompt

New [role] starts on [date], reporting to [name].

Their first three months should get them to:

[DESCRIBE WHAT "UP TO SPEED" MEANS]

Here is who they will work with and what our systems are:

[LIST]

Produce a 30/60/90 plan.

Format: week-by-week for the first month, then fortnightly.
Each item has an owner and a completion signal.

Include:
- Who they should meet in week one, and why
- What they should be able to do unsupervised by day 30
- A check-in at day 14 with specific questions to ask them

The day-14 check-in

Most new-starter problems are visible in week two and get raised in month three, by which point they have hardened. Scheduling the conversation in advance, with specific questions, is one of the cheapest retention interventions available.

Policies and difficult conversations

Strong fits, with the same caveat each time — a person owns the outcome:

Drafting handbook sections and policies from your rough rules.
Preparing for a difficult conversation: what to say, what to avoid, what they might say back.
Rewriting a policy so it is comprehensible to the people it applies to.
Summarising employment legislation as background reading — then confirming it with someone qualified, because this is exactly the kind of specific, changeable, jurisdiction-dependent fact AI gets confidently wrong.

Never paste an employee grievance or performance record

These are among the most sensitive records a business holds, and pasting one into a consumer AI account is a serious exposure. On a business tier with the right terms it can be defensible — but this is the category where you check the terms before, not after.

Can we use it to screen CVs at all?

Extracting structured information — years of experience, whether a certification is present — is a defensible clerical use, provided a person makes every decision and you can explain the process. Ranking or scoring candidates is not, and in several jurisdictions is explicitly regulated. If in doubt, get advice before, not after.

What about candidates using AI on their applications?

Assume they are, because they are. Cover letters have lost most of their signal. The response is to weight structured interviews and work samples more heavily, which are better predictors anyway.

Can it help write a performance review?

It can structure and improve the clarity of a review a manager has written, working from the manager's own observations. It must not generate the assessment. The distinction matters legally, and it matters to the person being reviewed.

Key takeaways

Never rank, score, or reject candidates with AI. That line is legal, not stylistic.
Start job adverts from the real work, and let it challenge your requirements.
Structured interviews with a rubric written in advance — AI makes this cheap enough to actually do.
A 30/60/90 plan with a scheduled day-14 check-in is one of the cheapest retention wins there is.
Never paste grievances or performance records into a consumer account.