Can ChatGPT make a staff rota? We tested it

ChatGPT can make a staff schedule, but you'll need to check it yourself afterwards. In a Shiftbase test in November 2026, ChatGPT only found the real problem in the week after we steered it twice. The schedules it made after that left 6 to 11 shifts open where 4 were unavoidable, and two of the three gave someone just 7.5 hours' rest after a closing shift.

It's still useful for a first draft, as long as you know where it goes wrong. The answers themselves show that best.

The test: a café week with nine staff

We asked ChatGPT what a café owner would ask: make the schedule for Monday 16 to Sunday 22 November. Every day there's a morning shift from 07:30 to 13:30 with a barista, someone in the kitchen and someone front of house. The afternoon shift from 13:00 to 18:00 needs a barista and someone front of house. On Friday and Saturday there's also an evening shift until midnight, with one person in the kitchen and one front of house, and one of them needs a key to lock up.

We ran the test for a café team in the Netherlands, with the prompts in Dutch. The rest rule in the prompt, 11 hours, matches the daily rest entitlement for adult workers in the UK. This is the team, with a ninth person who joined later in the test:

We spelled out the rules word for word: nobody works two shifts that overlap, there are at least 11 hours' rest between shifts, contracted staff don't go over their contract hours and zero-hours staff don't go over their maximum. We asked each question three times in the ChatGPT app and checked every answer with a script that tests exactly those rules.

The week has three tricky spots, like almost every real week. On Wednesday, Tim is the only barista, while the two barista shifts overlap by half an hour. Only Mehmet has a key, so he locks up on Friday and Saturday and can't open the next morning. And Tim is on annual leave from Friday to Sunday.

Round 1: ChatGPT asks for more hours

All three answers started with a sum. The week needs 222 hours and the team has 214 contract and zero-hours hours between them, so it's 8 hours short. ChatGPT then asked how we wanted to solve that: allow overtime, bring in another zero-hours worker or put fewer people on each shift.

The sum works on paper and misses the point. Tim's 28 hours counted in full, but with his leave he can work at most 24 hours across four days at one shift a day. The real bottleneck was Wednesday, and more hours don't fix that. One answer even suggested a zero-hours worker for kitchen and front of house: exactly the kind of person who leaves the barista gap open.

Round 2: with an extra person, it finds the real problem

We did what a manager would do next. We added Kim, a zero-hours worker with a maximum of 10 hours for front of house and kitchen who can work any day, and asked ChatGPT to make the schedule straight away. This time all three answers found the Wednesday problem: Sanne can't work, Daan only works weekends and Tim can't cover two overlapping shifts.

But no schedule came back. All three times, ChatGPT refused to make one while that single shift couldn't be filled. Roughly translated, it said it didn't want to hand over a schedule that looks complete but won't work in practice. That's honest. If you need the week sorted by Sunday evening, though, it doesn't get you far: in real life you leave that shift open and find someone for it.

Round 3: the schedule itself, checked

In the third round, ChatGPT was allowed to leave a shift open if nobody fitted. That produced three full schedules, each noting that it had followed the rules. Two of them said, word for word, that there were at least 11 hours' rest between shifts. This is what our script found:

We worked out the best schedule in the bottom row with a solver, a program that checks every combination. With this team and these rules, at least 4 shifts have to stay open. ChatGPT left 6 to 11.

The rest error is the most serious. Someone who locks up at midnight on Saturday and opens at 07:30 on Sunday gets 7.5 hours' rest. Under the Working Time Regulations, an adult worker is entitled to at least 11 consecutive hours' rest in each 24-hour period (regulation 10). For a closing shift followed by an opening shift, 7.5 hours falls well short. What else the rules say about rest and working hours is on the Working Time Directive UK page, and the clopening shifts page covers this exact pattern.

The open shifts are less obvious but cost you just as much. In attempt 3, the kitchen stayed empty on Friday evening, even though Joris could work that night and had 14 of his 20 contract hours left. As the manager, you'd have to spot that yourself by laying the whole schedule next to your team's availability. That's exactly the work you wanted to hand off.

Why a chatbot gets this wrong

ChatGPT writes the most likely answer to your question. It doesn't work through every combination step by step. This week had 39 slots to share between nine people, with rest, contract hours, skills and a key as conditions. A schedule that looks right is quick to write. A schedule that is right needs every rule checked for every combination.

On top of that, ChatGPT only knows what you paste in. Who's off sick, who swapped a shift and who's on leave next week: you type it all in again every week. And if you paste names, contract hours and leave into a chatbot, you're sharing staff data with an outside service. Check what your business has agreed on that, or use first names or initials.

Making a rota with AI: how to get more out of it

A chatbot still saves time on a first draft, especially for a standard week you then reuse as a template. These things made the difference in our test:

• Put everything in your question: every shift with its start and end time, plus each person's contract hours, what they can do and which days they can't work.

• Write out the rules, even the obvious ones: 11 hours' rest, no overlapping shifts, who can lock up.

• Let a shift stay open. Without that line we didn't get a single schedule; with it we got three.

• Ask for a table with one row per role per shift, so you can see straight away which slot is empty on which day.

• Check it yourself: the rest after every late shift, each person's hours and whether every role is covered. Two of the three schedules said the rest rules were met, and someone still got 7.5 hours after a closing shift.

Treat the result as a draft. Copy it into your own schedule or a schedule template in Excel and check it there. If the gaps are what slows you down, the open shifts page explains how to let your team pick them up.

How Shiftbase auto-scheduling is different

Auto-scheduling in Shiftbase works with the data that's already in Shiftbase: the shifts you need, the availability your team shares, contract hours, skills and the hour limits in your settings. Shiftbase then fills the shifts in one go, across several departments at once if you want.

Whatever can't be filled stays visible. You see for each person why they weren't picked, and you post the remaining shifts as open shifts your team can claim in the app. Shiftbase never publishes for you: you review the draft with the labour cost next to it, change what you want and publish it yourself. If you'd rather schedule by hand, that works as it always has.

At WONDR Experience in Amsterdam, the weekly schedule for five departments went from two full working days to two hours. See how it works with several departments and your labour cost in view on the auto-scheduling page. You can try Shiftbase free for 14 days with your own team, no payment details needed.