Claude for
people work
Policies, job descriptions, interview guides, sensitive communications, and the places where tone matters as much as content.
HR is one of the places where Claude's help is most valuable and also the place where you have to be most careful about what goes in. The enterprise account protects your data from training, but names, compensation, performance details, personal circumstances, all of that deserves real thought before it becomes a Claude prompt. This page assumes you're going to use judgement on that. The workflows below are the ones where Claude genuinely earns its keep.
Three places to start
Three workflows where Claude does a real share of the drafting without getting in the way of the human judgement that matters most.
1. Refining a job description
Most JDs are written once and then copy-pasted forever. Claude is excellent at tightening them: sharpening the first two sentences so the right candidate wants to read on, cutting generic filler, and trimming the requirements to the three or four that actually matter. Give it the current draft and a JD you were happy with as a voice reference.
JD refinement prompt, annotated
Four elements applied to the "make this JD better" ask.
Where this falls over
Claude will tone down anything that sounds edgy or opinionated, which is often the thing that would attract the right person. If your reference JD has personality, tell it explicitly to preserve the voice.
2. Building a structured interview guide
Paste the JD and ask for a 45-minute interview guide: intro, behavioural questions with follow-ups, a small practical task, wrap. For each behavioural question, ask Claude to explain what it's trying to learn. The extra explanation is what lets you tell which questions are actually useful and which are generic.
Interview questions, weak vs useful
Same role, two prompts. The second gives you questions you can actually defend.
"Tell me about yourself." "Describe a challenging project." "What is your greatest weakness?" "Where do you see yourself in five years?" Generic questions that tell you nothing about whether this person can actually do the job.
A 45-minute structured guide. Three behavioural questions, each with a one-line "what this is testing" annotation:
- Tell me about a time you had to make a call with incomplete data. Testing: can they tolerate ambiguity and still commit to a defensible position.
- Walk me through a disagreement with a product manager where you were right and they were wrong. Testing: do they push back constructively, or avoid conflict, or get combative.
- Show me a chart or report you made recently that you would now simplify. Walk me through what you would cut and why. Testing: do they have editorial judgement, or do they treat every finding as equally important.
Plus a 10-minute practical task: a messy CSV of quarterly metrics and a two-sentence brief from a fictional PM asking for "insights". See what they do with it.
Where this falls over
Without the "explain what each question is trying to learn" instruction, the output is generic interview questions you could have got from a template. That instruction is what makes it worth doing.
3. Drafting sensitive employee communications
Layoffs, policy changes, sensitive announcements, difficult emails. These are the messages where tone and content both matter, and Claude is genuinely good at both. Give it the key points, the constraints on what can and cannot be said, and the audience. The output is a first draft, not a final version, but a first draft saves more than half the work.
Drafting a tricky message, as a conversation
Notice that Claude asks the clarifying questions before it writes anything.
Where this falls over
Claude's default is warm but slightly bland. The fix is to give it explicit instructions on tone ("warm but clear, no corporate hedging, no false cheer") and to push back on anything that sounds like canned corporate language.
Worked examples
Before and after, on tasks from HR work.
Prompts to steal
Copy any of these, fill in the square brackets, run.
Try this week
Pick one. All of these are under an hour.
The boundary for HR is sharper than for most roles: no personal data, no performance specifics, no compensation numbers tied to named people. Everything above that line is fair game, and the upside is significant.
If you want to go broader, the Beginner guide covers the habits that apply everywhere. The prompt library has the full set of role-tagged prompts in one place.