Guide
Custom instructions for HR: what goes in, what stays out
HR custom instructions should hold what's true in every chat: your role and company, the tone for employees and candidates, and the rules that protect people. Those rules are job-related criteria only, no guesses about anyone's age, health, family or background, and a check mark on anything about pay, leave, discipline or the law. Keep people's data out: CVs, interview notes and employee records belong in your HR system or a project on a business plan, not in instructions every chat reads, and the tasks you repeat, such as interview scorecards, go into skills so every interviewer rates on the same anchors. Keep it short; OpenAI's help page, read through search-result text, gives 1,500 characters on Free and Go and 5,000 on Plus and above.
Short answer
- In: role, tone, job-related criteria only.
- In: a check mark on pay, leave, discipline and the law.
- Out: CVs, notes, employee records.
- Repeated tasks become skills.
What goes in, and what stays out
| Line | Where it goes |
|---|---|
| Your role, company size, tone for employees and candidates | Custom instructions |
| Job-related criteria only; no guesses about protected traits | Custom instructions |
| Check marks for pay, leave, discipline, termination, the law | Custom instructions |
| The handbook and policies | A file in an HR project, quoted when needed |
| Candidates' CVs, interview notes, employee records | Your HR system, or a project on a business plan; never custom instructions |
| Interview scorecards, job posts, policy FAQs | Skills |
A worked example
Illustration: a made-up business. The HR coordinator at a 45-person US software company had a salary band for each open role and two candidates' names in her custom instructions, left over from last quarter's hiring. The bands moved to the HR project and the names were deleted. What stays, at 560 characters:
Custom instructions: HR coordinator
I'm the HR coordinator at a 45-person US software company. Plain, kind words for employees and candidates; short sentences.
Judge people only on job-related criteria from the role's scorecard. Never mention or guess age, health, family, religion, national origin or other protected traits, and don't use "culture fit".
Mark anything about pay, leave, discipline or termination [check: HR policy], and anything legal [check: counsel].
Our policies are in my HR project; quote them, don't paraphrase from memory.
For interview scorecards, use my scorecard skill.
The next time a hiring manager asked her to "summarize which candidate seems like the best fit", the draft compared the two finalists against the scorecard's four competencies and left "fit" out.
Next step: bring the setup to onboarding
The same split, instructions for every chat and skills for repeated work, carries into a new hire's first weeks. AI for onboarding new employees: a manager's guide covers the plan and the setup a new hire gets.
Skills for this
- HiringStructured interview scorecardBuilds a structured interview scorecard: 4 to 6 job-related competencies, 1 to 4 behavioral anchors, evidence fields and a hire/no-hire rule set in advance.
- HiringJob post from a role intakeWrites a job post with an honest title, the first 6 months of work, 5 or fewer must-haves, a pay-range line checked for local rules, and an inclusive-language check.
- HiringCandidate summary for a debriefSummarizes interview notes, scorecards and CV into a debrief brief: evidence per competency, disagreements, concerns, open questions, and a separate recommendation.
- AI practiceContext brief for an AI assistantWrites instructions for an AI assistant or agent: role, audience, can and can't, sources of truth, tools, format, examples, edge cases, and handoff.
Questions
What should HR put in ChatGPT custom instructions?
What holds in every chat: your role and company size, the tone for employees and candidates, judging people only on job-related criteria, never guessing about protected traits such as age, religion or national origin, and a check mark on anything about pay, leave, discipline or the law. Point to where your policies live instead of pasting them.
Can I paste CVs or employee records into ChatGPT?
Not into personal accounts: on personal Free, Plus and Pro accounts, training on chats is on by default. Use your HR system or a project on a business plan your company controls, paste only what the task needs, and never put people's data in custom instructions, which every chat reads.
How do I make interview ratings consistent with AI?
Put the method in a skill, not in custom instructions: a structured interview scorecard with job-related competencies, anchors for each rating and a hire rule set in advance. The US Office of Personnel Management describes structured interviews as measuring job-related competencies by asking every candidate systematically. Each interviewer's AI then rates on the same sheet.
AI skill
Have your AI build the scorecard for your interviews
Put this skill in AI Skill Pod and everyone on your team gets it in their Claude, Cursor or ChatGPT (Pro, Business, Enterprise): one invite link, free to start. With Structured interview scorecard, each interviewer's AI builds the scorecard your instructions point to: four to six job-related competencies, behavioral anchors for each rating, fields for what the candidate actually said, and a hire rule written before the first interview.
14 days free · No card needed
Sources
- OpenAI Help: ChatGPT custom instructions — up to 1,500 characters on Free and Go, 5,000 on Plus, Pro, Business, Enterprise and Education; read through search-result text, the page blocks automated reads (read 2026-11-05)
- EEOC: Prohibited employment policies and practices — illegal to discriminate against an applicant because of race, color, religion, sex, national origin, age (40 or older), disability or genetic information (read 2026-11-05)
- US Office of Personnel Management: Structured interviews — an assessment designed to measure job-related competencies by asking systematically about past behavior or hypothetical situations (read 2026-11-05)
- OpenAI Help: Data controls — on personal Free, Plus and Pro accounts, training on chats is on by default and can be turned off; read through search-result text (read 2026-11-05)