LunaPrompts
HR can use AI to prepare job descriptions, organize hiring requirements, draft candidate messages, explain approved policies, and build onboarding plans. Start with work that a colleague can review before anyone acts on it. A useful first project has a clear input, a recognizable output, and an owner who can tell whether it is right.
Picture a Monday morning with three open roles, a new starter, and a manager asking for a hiring update. Buying another tool is not the first decision. Choosing which of those tasks deserves help is.
This guide gives you ten places to start, a worked example, and a small pilot you can run with your existing approved tools. If you already know the task, go straight to the 15 ChatGPT prompts for HR.
Where AI fits in an HR team
Think about three kinds of work. Preparation creates a draft: an email, a question set, a checklist. Coordination moves agreed information between people and systems. Evaluation influences what happens to a person. Each needs a different level of review.
Start with preparation. A missing sentence in a draft is easy to correct. A mistaken rejection, a changed employment record, or an incorrect answer about an employee’s circumstances is harder to undo. Keep those decisions with the responsible person and the approved process.
The workflows below are practical suggestions, not measured claims about time savings.
10 practical ways HR can use AI
| Workflow | Useful output | Who checks it |
|---|---|---|
| Hiring intake | A role brief and missing questions | Hiring manager |
| Job descriptions | A readable draft tied to real work | Recruiter and manager |
| Candidate communication | Invitations and process FAQs | Recruiting owner |
| Interview preparation | Questions mapped to skills | Interview panel |
| Assessment design | Task and draft scoring rubric | Subject-matter expert |
| Onboarding | First-week plan with owners | Manager and operations |
| Policy communication | FAQ from approved text | Policy owner |
| Learning materials | Workshop and practice exercises | Trainer |
| Recruiting reporting | Explanation of aggregate counts | Analyst or recruiter |
| Vendor evaluation | Demo script and pilot checklist | Buying team |
1. Turn a messy intake into a clear role brief
Give the assistant approved notes about the work, the team, and the first three months. Ask it to separate essential skills from skills someone can learn. Require a list of missing information instead of allowing it to fill gaps.
For example, “comfortable with AI” is too broad to assess. “Can build a support assistant, test unsupported answers, and explain failure handling” gives the recruiter and manager something concrete to discuss. The manager still needs to approve that interpretation.
2. Write job descriptions people can understand
Use the role brief as the only source of requirements. Ask for a short role purpose, the work someone will own, essential skills, and the hiring process. Review every claim about pay, benefits, location, and flexibility. A fluent draft can still promise something the company never agreed to offer.
Google documents job-description drafting in Docs as one HR use case for Workspace with Gemini. The broader workflow can also be tried with another company-approved writing assistant. Google’s HR collaboration guide.
3. Make candidate messages more useful
An assessment invitation should answer practical questions: how long it takes, what to submit, which tools are allowed, what happens next, and where to request help. Ask AI to check a draft against that list. It may be more useful as an editor than as the first writer.
Keep the recruiter responsible for recipients, dates, and commitments. Review before sending.
4. Prepare interviews around evidence
Ask for questions linked to the role’s actual tasks. Add a neutral follow-up and the evidence each question is intended to reveal. Give the panel time to remove duplicates and agree difficulty.
A question such as “tell us about AI” leaves too much room for rehearsed vocabulary. “Show how you would check this assistant’s answer against its source” produces something the interviewer can inspect. Use the free interview scorecard builder to organize agreed criteria.
5. Draft a practical assessment
AI can help turn a role brief into a task, deliverables, and a draft rubric. Then ask a subject-matter expert to complete the task and challenge the scoring criteria. Check whether the expected time and tooling are realistic.
For AI engineering roles, start with our assessment templates. LunaPrompts is designed for the next step: running practical AI and technical skills assessments and reviewing candidate evidence. A draft produced in a chat window is a starting point for that process.
6. Coordinate the first week
Create an onboarding table with an activity, owner, dependency, and completion check for each day. Use role placeholders while planning. Flag unassigned work explicitly: “IT owner missing” is more useful than an invented name.
OpenAI’s onboarding example uses approved context to prepare trackers and welcome drafts for review. Connected actions depend on the tools and access available in the workspace. OpenAI’s onboarding workflow.
7. Explain an approved policy
Provide the policy version and effective date. Ask for a plain-language FAQ with a section reference for each answer. Try a question the policy does not answer and see whether the assistant acknowledges the gap.
A generic model answer is not your company’s policy. Have the policy owner review the draft, especially where exceptions or individual circumstances matter.
8. Create training people can practice
Ask for a short workshop with a fictional scenario, a task, and a review exercise. For AI literacy, deliberately include an invented fact in a sample output and ask participants to find it. This teaches checking, not just prompting.
Keep a small set of examples showing an acceptable result, an incomplete result, and a result that should be escalated.
9. Explain recruiting numbers
Use aggregate stage counts with consistent dates and stage definitions. Ask for numerators and denominators alongside every percentage. Recalculate in a spreadsheet before sharing.
If an illustrative cohort has 80 completed assessments and 20 interview invitations, the invitation rate is 25%. That number alone does not explain whether the assessment is useful. Task difficulty, sourcing, role requirements, and the review process can all matter.
10. Make vendor demos specific
Give each vendor the same workflow and ask them to show it. Include an incomplete input, an output correction, and an export. Record what you saw, what was promised, and what still needs checking. The 2026 AI tools shortlist helps match tool categories to these tasks.
A worked example: from role intake to interview plan
Suppose a team is hiring an engineer to build an internal support assistant. The manager’s notes say “Python, RAG, good communication.” Ask the assistant to turn those notes into questions: Which documents will be searched? What should happen when there is no supporting evidence? Who will review the answers?
After the manager answers, draft a task using a small fictional handbook. Ask the candidate to return an answer with supporting passages and to handle an unanswerable question. A reviewer can then inspect grounding, failure behavior, and explanation.
The interview follows the submitted work: why did the candidate choose that retrieval method, and what failed in testing? Our RAG assessment template provides a starting rubric. This is an illustrative workflow; calibrate it for the role before using it to select candidates.
Run a two-week pilot
Before starting: choose one recurring task, its owner, and the approved inputs. Collect a small baseline of comparable tasks. Record time spent drafting, reviewing, and correcting.
During the pilot: use the same output checklist for manual and AI-assisted work. Keep examples of mistakes and successful corrections. Ask the reviewer whether the output is easier to use, not only faster to produce.
At the end: compare total time, material corrections, and completion rate. Write down limits such as a small sample or unusually simple tasks. Expand only when the improvement survives review. Use the 2027 planning guide to turn that pilot into a broader plan.
Questions HR teams ask
Does HR need coding skills to use AI?
Not for drafting and reviewing text with an approved assistant. Connecting systems or automating actions may require administrators or technical support. Start with a task you can inspect manually.
Should we paste employee records into a chatbot?
Use only data and workspaces approved by your organization for the task. Prefer fictional or minimized examples during a pilot. Verify access, retention, and sharing before using sensitive records.
How do we know whether AI is helping?
Measure the complete workflow, including review and corrections. Faster first drafts are useful only if the final result is accurate enough for the intended use.