Claude for HR: A Practical Guide
How HR teams can use Claude for policy Q&A, review cycle setup, and survey analysis. Covers Claude Chat, Cowork, skills, and connectors with real examples.
Most HR teams have tried ChatGPT for a one-off draft. Far fewer have used Claude to build a policy chatbot, spin up a review cycle, or generate an interactive dashboard from survey data. This guide walks through the setup for each of those workflows, using the Windmill webinar below as the reference.
The examples below assume no prior experience with Claude. You can follow them with a Claude Pro or Team subscription and access to your existing work tools.
The three ways to run Claude
Claude comes in three flavors, and the one you pick determines which workflows are possible. Claude Chat runs in the browser, like ChatGPT. Claude Cowork and Claude Code run locally on your computer, so they can read files on your machine and connect to your other apps.
For HR work, Claude Cowork is the sweet spot. Claude Chat is fine for simple questions. Claude Code might be overkill unless you’re writing software (though some HR teams do use it for scripting and data work). Cowork sits in the middle — it can analyze a CSV on your desktop, pull from Notion, and send a Slack message in a single conversation.
All three share the same underlying models. The differences are about what the AI can reach, not how smart it is.
Skills and connectors, explained
Two terms dominate the Claude vocabulary, and they’re easy to confuse.
Skills are reusable prompts. If you find yourself pasting the same instructions every time you analyze a pulse survey, save those instructions as a skill and invoke it with a slash command. Under the hood, a skill is a Markdown file — nothing more.
Connectors are integrations with other apps: Notion, Slack, GitHub, Gamma, your HRIS. When you authenticate a connector, Claude inherits your permissions in that app. It cannot see Notion pages you can’t see, and it cannot send Slack messages in channels you’re not in. This is the foundation of the Claude security model.
You can see both in Claude Cowork under Customize. The connector settings also let you block individual actions — for example, blocking Slack’s “send message” tool while still allowing Claude to read messages.
Use case 1: A policy Q&A bot
HR teams field the same policy questions every week: vacation balance, parental leave, expense limits. Claude Cowork can answer these from a single source of truth.
The simple version: drop a PDF of your policies into a folder, open Cowork in that folder, and ask questions. Cowork will find the PDF and answer from it. The problem is that PDFs go stale the moment a policy changes.
The better version uses Notion (or any connected knowledge base) as the source. Build a small skill that tells Claude: “If someone asks an HR question, search the Notion policies database and answer from there.” Now when you update the Notion page, every employee using the skill gets the new answer on their next question. No resending files, no version drift.
For broader rollout, put the same logic behind a Slack bot so employees don’t need Claude access at all. Windmill’s Slack-native agent handles similar workflows out of the box.
Use case 2: Spinning up a review cycle
Setting up a performance review is usually a copy-paste slog: open the tool, paste questions, pick dates, configure windows, write a launch announcement. Claude can collapse the whole thing into a conversation.
Start with clarifying questions
The single most useful prompting trick for HR tasks: tell Claude to interview you first. Instead of asking “write me three self-review questions,” ask “write me three self-review questions, but first ask me four clarifying questions about my company, review philosophy, and audience.”
You’ll get much sharper output because Claude pulls context from you before writing. Generic self-review questions become questions tailored to your values, team structure, and review goals.
Connect your tools
Once you have the questions, don’t paste them manually. If your review platform has an MCP connector (Windmill does), tell Claude to create the cycle directly: “Use the Windmill connector to create this cycle with the questions above, a 5-day self-review window, and a 3-day manager review window.”
Generate the rollout deck
From the same conversation, ask Claude to create a Gamma presentation explaining the cycle to the company. Include a link to your review platform’s help docs in the prompt so the deck explains how the tool works, not just what the schedule is. The output won’t be perfectly on-brand, but it will get you to 80% in minutes instead of hours.
Use case 3: Survey data analysis
A pulse survey CSV is best analyzed in three steps. Most HR teams stop after the first two — which is exactly where Claude Cowork starts to pay off.
Step 1: understand the shape. Drop the CSV into a folder and ask Cowork what’s in it. You’ll get a summary of columns, response counts, and initial patterns. This catches data issues before you waste time analyzing bad inputs.
Step 2: structured analysis. Ask for trends, attrition risk scoring, open-ended theme extraction. Open-ended questions used to be painful because no one wanted to read 200 free-text responses. Claude reads them in seconds and clusters the themes. Ask for weighted scores (Claude will write Python, show you the weights, and let you adjust them) to turn unstructured feedback into numeric signals.
Step 3: build a dashboard. Here’s the leap most teams miss. Ask Claude to generate an interactive HTML dashboard from the data — spider charts by department, drill-downs by tenure, filterable employee lists. One prompt produces a standalone HTML file you can share directly or deploy to Vercel with authentication for the exec team.
The web artifacts skill (built into Claude) is specifically designed for this. Enable it under Skills → Built-in before you try.
Practical guardrails
Three things to get right before rolling Claude out broadly.
Approved tools, not policed access. The worst-case scenario is locking Claude down so hard that employees use personal ChatGPT accounts for work data. Pick a sanctioned set of tools, turn on zero-data retention, and let people use them freely. Centralized sanctioned access beats fragmented shadow AI every time.
Humans make the decisions. Use AI to gather context, find patterns, and draft narratives. Never delegate hiring, promotion, rating, or termination decisions to AI. Every major employment AI regulation (NYC Local Law 144, the EU AI Act, Illinois AB 3773) targets fully automated decisions, so keeping humans in the loop is both ethical and legally safer. Windmill’s approach to calibrations is a good reference point — AI surfaces bias and generates pre-reads, but managers make the calls.
Watch for AI slop. It’s easy to generate a 40-slide deck in two minutes. That doesn’t mean it’s good, or that anyone should read it. Require humans to review AI output before it goes anywhere external. If someone sends you something they clearly never read, send it back.
Where to go next
The workflows above are the on-ramps. Once your team is comfortable with skills and connectors, the real leverage comes from scheduling them to run automatically — pulling survey results every Friday, pinging managers when new hires start, generating calibration pre-reads before review cycles. That’s covered in the advanced workflows guide.
For HR-specific automation that goes beyond what you can build in Claude alone, Windmill handles the integration layer, security model, and agent orchestration end-to-end.
Frequently Asked Questions
What is Claude and how do HR teams use it?
Claude is Anthropic's AI assistant. HR teams use it for policy Q&A, drafting performance review questions, analyzing survey results, and generating presentations. Claude runs through three products: Claude Chat (web-only), Claude Cowork (local file access), and Claude Code (for engineers).
What's the difference between Claude Chat and Claude Cowork?
Claude Chat runs in the browser and cannot access your computer. Claude Cowork runs locally, so it can read files on your machine, connect to apps like Notion and Slack, and invoke skills. For most HR workflows involving documents or multiple tools, Cowork is the better choice.
What is a Claude skill?
A skill is a reusable prompt saved as a text file. Instead of pasting the same instructions every time, you invoke the skill with a slash command. Skills are just Markdown — they contain no secret data — and can be shared with colleagues as zip files or through organizational plugins.
Is it safe to put HR data into Claude?
Claude offers zero-data retention on enterprise plans, meaning Anthropic does not save your logs or train on your data. Get IT approval for your preferred tools, then treat Claude like any other vendor your organization has vetted. The bigger risk is employees using personal AI accounts outside sanctioned tools.
Should AI make performance decisions?
No. Use AI to gather context, draft narratives, and surface patterns, but keep humans in charge of hiring, promotions, ratings, and terminations. Most emerging regulation (NYC Local Law 144, EU AI Act) specifically targets fully automated employment decisions, so human review is both the ethical and legally safer approach.