ChatGPT prompts for staff rostering

The difference between a useless roster and a genuinely helpful draft is almost always the prompt. Think of it as a brief you’d give a new duty manager: the more specific you are, the less you have to fix afterwards. Every prompt in this guide is built from the same five ingredients.

The fifth ingredient is the output format — spell out how you want the answer. “Return a Markdown table, one row per person, columns Monday to Sunday, with a total-hours column” gets you something you can paste straight into a spreadsheet. Leave it out and you’ll get prose you have to reformat by hand.

Keep each prompt focused on one job. A single mega-prompt that asks ChatGPT to build, optimise and cost a roster at once tends to do all three badly. The prompts below split those jobs up.

This is the workhorse. Notice how it names a persona, lists people by role, states the shifts to cover, and ends with an explicit output format. Replace the team and shifts with your own — the structure is what matters.

You are an experienced venue manager building a weekly staff roster for a busy café.

Team and availability (use these role labels, not real names):
- Supervisor A — available Mon–Fri, not weekends
- Barista B — available all week, max 4 days
- Barista C — available Wed–Sun
- All-rounder D — available weekends only
- Casual E — available any day, wants 2–3 shifts

Shifts to cover every day:
- Open 6:00–14:00 — needs 1 supervisor + 1 barista
- Close 14:00–22:00 — needs 2 staff

Rules:
- No one works more than 38 hours or 5 days per week
- At least one supervisor on every Open shift
- No "clopening" (a Close shift followed by the next morning's Open for the same person)
- Rotate weekend shifts fairly across the people available on weekends
- Everyone gets at least two consecutive days off

Output: a Markdown table with one row per person and columns Mon–Sun showing the
assigned shift or OFF. Add a final column with each person's total hours for the week.

ChatGPT returns a completed grid you can paste into Excel or Google Sheets and refine. If something looks off, don’t rewrite it manually — reply in the same chat with the fix (“Barista B can’t do Thursday, redo just that day”) and let it adjust.

Once a base roster exists, ChatGPT is great at evening it out. Paste your current grid in and ask it to redistribute — this is far faster than dragging cells around and recounting hours.

Here is my current roster as a table:
[paste your roster table here]

Rebalance it so that:
- Total hours are spread as evenly as possible across all staff
- No one works more than two weekends in a row
- Late finishes (after 22:00) are shared fairly, not always the same people
- Coverage stays exactly the same — don't leave any shift short

Keep everyone within their stated availability and their 38-hour limit.
Show the updated table, then below it list exactly what you changed and why.

The “list what you changed and why” line is the important bit — it turns a black-box reshuffle into something you can actually check before you publish. Fairness in how shifts and penalties are spread is one of the biggest drivers of staff satisfaction, so it’s worth being explicit about it.

The most stressful rostering moment is a same-day sick call. ChatGPT can shortlist replacements in seconds — provided you give it the roster and the constraints so it doesn’t suggest someone who’s already maxed out.

Barista C has called in sick for today's Open shift (6:00–14:00, needs 1 barista).

From this roster and availability list:
[paste roster + availability here]

Suggest up to three replacements. For each person, tell me:
- Whether they're already rostered today, and if swapping breaks their hours or days limit
- Whether they're qualified for the barista role
- Whether calling them in would create a clopening or a 6th day in a row

Rank the options from least disruptive to most, and note who to avoid and why.

You still make the call — but instead of scanning the whole roster in a panic, you get a ranked shortlist with the trade-offs spelled out.

Before you publish, get ChatGPT to play devil’s advocate. This catches the copy-paste mistakes that are easy to miss when you’ve been staring at the same grid for an hour.

Review this roster for problems before I publish it:
[paste your roster table here]

Check for and list every instance of:
- Anyone double-booked (two shifts at the same time)
- Anyone over 38 hours or more than 5 days
- Any shift left uncovered or missing a required supervisor
- Long stretches (6 or more days in a row) or clopenings
- Uneven weekend or late-shift distribution

For each issue, name the person or shift and suggest a fix.
If a category has no issues, say so explicitly.

Think of this as a second pair of eyes, not the final word. ChatGPT will catch structural problems like double-books and over-hours, but it can’t see fatigue history, team dynamics, or your award obligations — which is exactly where the next prompt comes in.

This is the prompt most people skip — and the one that keeps you out of trouble. ChatGPT does not have reliable, current knowledge of Australian Modern Award rates, penalty loadings, overtime thresholds or minimum-engagement rules, and those figures change every year. If you ask it to “cost this roster”, it will often produce confident numbers that are simply wrong.

So don’t ask it to calculate pay. Ask it to flag anything that affects pay, so you can verify against the actual award:

Important: you do NOT have access to current Australian Modern Award pay rates,
penalty rates, or overtime rules, and these change regularly. Do not calculate,
estimate, or invent any pay figures or dollar amounts.

Instead, when you build or review this roster, FLAG anything that would affect pay
so I can check it against the correct award and Fair Work rules — for example:
- Shifts on evenings, weekends or public holidays (penalty rates likely apply)
- Anyone crossing from ordinary hours into overtime
- Shifts that may trigger minimum-engagement periods or paid breaks
- Junior, casual or apprentice staff whose rates differ

List these as "assumptions to verify", not as final costs.

This turns ChatGPT from a source of made-up numbers into a useful checklist of things to confirm. It’s the honest division of labour: the AI spots where pay rules might apply; you (or your rostering software) apply the actual rates.

The prompt library above will genuinely save you time on the creative part of rostering — laying out shifts, balancing hours, backfilling gaps. What it can’t do is the compliance part, and that’s not a prompt-writing problem you can solve with a cleverer instruction. ChatGPT has no live award data and no connection to your payroll.

That’s the gap purpose-built software fills. RosterElf’s automatic award interpretation applies the correct pay rates, penalties and overtime as you build the roster — so the cost you see is the cost you’ll pay, not an AI guess. Pair that with the pattern-spotting speed of AI-assisted rostering and you get the best of both: fast drafts and correct pay.

If you want to keep working in a spreadsheet for now, two resources bridge the gap between a ChatGPT draft and real numbers:

Prefer a different AI, or a spreadsheet-first workflow? The same prompt principles carry across. See how to create a staff roster with Claude, build a roster in Google Sheets with ChatGPT, or create a staff rota in Excel using ChatGPT. The persona-context-data-rules-output recipe works in all of them.

ChatGPT is a genuinely useful rostering assistant when you brief it well — a clear persona, real data, hard rules and a defined output format turn it from a novelty into a time-saver. Just remember where its knowledge ends: it drafts and rebalances, but it doesn’t know your award, and it will happily invent pay figures if you let it. Use the prompts for speed, verify against Fair Work, and lean on software for the compliance and payroll that no prompt can deliver.