Try 3 lessons from AI for HR Professionals

Tools like ChatGPT, Claude and Gemini are built on large language models (LLMs). An LLM has been trained on an enormous amount of text and has learned to predict what words are likely to come next. When you ask it a question, it writes the answer one small piece at a time, each time choosing what fits best given everything before it.

That one idea explains most of what you'll experience at work:

• It is excellent with language. Drafting, rewriting, summarising, changing tone, structuring messy notes. These are prediction problems, and it is very good at them.
• It doesn't know your company. It has never seen your policies, your people or your culture unless you give them to it in the conversation.
• It can be confidently wrong. If the most plausible-sounding answer isn't true, it may still write it, with no change in tone. This is often called a "hallucination".
• It has a working memory limit. Each tool can only consider a certain amount of text at once, called its context window. Very long conversations or documents can push earlier details out.

Three kinds of "automation" HR teams meet

Generative AI is the most flexible of the three, and the one most likely to be used informally by your team today. That is why the rest of the course spends as much time on judgement and governance as on the tools themselves.

Most HR professionals who say "I tried ChatGPT and the answers were generic" were giving it generic prompts. The tool can only work with what you tell it. A good prompt briefs it the way you'd brief a capable new colleague, using five parts:

1. Role: who it should act as. "You are an experienced HR business partner in a mid-sized Indian IT services company."
2. Task: one clear verb. "Draft…", "Summarise…", "Compare…", "Critique…"
3. Context: what it can't know. The audience, the situation, what's been tried, the constraints.
4. Format: the shape of the answer. A table, five bullets, an email under 150 words.
5. Limits: tone, things to avoid, policies to respect, and "ask me if anything is unclear".

Bias-aware prompting

In HR, the way you ask matters as much as what you ask. A prompt like "Write a JD for a young, energetic sales executive" invites age-biased output. Ask for criteria tied to the work instead, and add a limit such as "Use inclusive language and avoid age- or gender-coded words."

Iterate, don't restart

Treat the first answer as a draft. A short, specific follow-up beats a new prompt: "Shorter, and lead with the date", "Too formal for Slack", or "List the assumptions you made."

You are [role].

Task: [one clear verb + what you need].

Context:
- Audience: [who will read it]
- Situation: [what is going on]
- Constraints: [policy, timing, sensitivities]

Format: [shape and length].

Limits: [tone, words to avoid]. Use inclusive language. If anything is unclear, ask me before you answer.

A job description is often the first thing a candidate reads about you, and small word choices change who applies. Research on job adverts has found that gender-coded language (words like "dominant" or "nurturing") can affect how appealing a role feels to different groups. Long lists of "requirements" that are really preferences can also put off capable people who don't tick every box.

AI is useful twice here: to draft the JD from a hiring manager's notes, and then to audit it, with you making the final call.

Step 1: Draft from the real job

Start with what the person will actually do in their first six months, not with an old JD. Give the AI the hiring manager's notes, the team context and your format.

Step 2: Audit it

Ask the AI to review the draft against a checklist:

• Gender-coded words (for example "aggressive", "rockstar", "nurturing")
• Age-coded phrases ("young and dynamic", "digital native", "recent graduate" when experience doesn't matter)
• Requirements that aren't really needed: years of experience, specific degrees, "native speaker"
• Jargon and internal acronyms an outsider wouldn't understand
• Missing information candidates care about: location, flexibility, how to apply, accommodations

Step 3: Decide

The AI will suggest changes. Some will be right; some will water down a genuine requirement. You and the hiring manager decide. The aim is a JD that describes the job accurately and invites everyone who could do it.

You are an inclusive-hiring specialist. Audit the job description below.

For each issue, quote the exact phrase, say why it may put off qualified candidates, and suggest a replacement. Check for:
1. Gender-coded words
2. Age-coded phrases
3. Requirements that may not be essential (years, degrees, language)
4. Jargon or unclear internal terms
5. Missing information candidates need

Format as a table: phrase | issue | suggestion.
Don't rewrite the whole JD yet.

Job description:
[paste JD]