Yes, AI can build the grid of a school timetable, but only the part a computer can prove: no teacher, room or class in two places at once. It cannot judge fairness, pupil welfare or what a particular teacher will tolerate. A person must still check part-time hours, room fit, option blocks and the awkward days.
The word "AI" covers two very different things here. A timetabling solver searches millions of arrangements against rules you type in and can show that none is broken. A chat assistant writes text that looks like a timetable and cannot prove that it is one. Most disappointments come from using the second where the first was needed.
The constraints, sorted by who must check them
Every timetable is a pile of constraints. Some are hard: if one breaks, the timetable cannot run. Others are soft: breaking them makes the week worse but not impossible. A solver treats the first kind as a wall and the second kind as a score to improve. Your job is to sort the list honestly before anything runs.
| Constraint | Hard or soft | Solver handles it? | Human still checks |
|---|---|---|---|
| A teacher is in one room at a time | Hard | Yes, reliably | Only that the staff list is right |
| A class is in one place at a time | Hard | Yes, reliably | Split groups and mixed-year classes entered correctly |
| Room capacity and room type (lab, studio, hall) | Hard | Yes, if you enter capacities | Rooms that are shared, closed for repairs or booked by outside hirers |
| Part-time teacher works only agreed days | Hard | Yes, if you enter the windows | Lunch, travel and "not first thing" requests nobody wrote down |
| Double periods for practical subjects | Hard or soft | Yes | Whether a double really survives a fire drill or assembly |
| Option blocks (pupils choose subjects that run at the same time) | Hard | Partly: needs the option data first | Pupils whose choices still collide |
| Heavy subjects not all in the afternoon | Soft | Scores it | Whether the score matches your view of a good day |
| No teacher with seven straight lessons | Soft | Scores it | Duty rotas, break supervision, trips |
| Fairness between staff (free periods, awkward slots) | Soft | Only if you define it | Always: this is a people question |
Notice what sits in the right-hand column: facts nobody entered. The solver is exact about the rules it was given and blind to the ones it was not. A good habit is to keep a running "unwritten rules" list through the year (the teacher who cannot do Friday afternoons, the minibus that leaves at 13:10 for Thursday matches) and type each one in at the start of planning.
Solver, chat assistant or both
A dedicated solver takes structured input: teachers, subjects, classes, rooms, lesson counts, availability. It returns a complete timetable or tells you it cannot. One free example is FET, open-source software for scheduling the timetable of a school, high school or university; its own site says it usually solves a complicated timetable in at most 5 to 20 minutes, a simpler one in under five (sometimes seconds), and an extremely difficult one can take hours. Commercial products exist too, and several now advertise AI features. Treat that word as a prompt to ask what the feature does: generate the grid, suggest a fix, answer staff questions, or write the explanation? Check the vendor's current page rather than a comparison site.
A chat assistant such as ChatGPT, Claude or Gemini is a language model. It is excellent at the work around the grid and unreliable at the grid itself, because it predicts plausible text rather than searching for a valid arrangement. Ask it for a five-day, six-period timetable for eight classes and it will return a tidy table. Check that table with a pencil and you will often find a lab used twice in period three.
The sensible split is a division of labour.
- Solver: builds the clash-free grid and re-runs it when staffing changes.
- Chat assistant: turns a messy staff email into a clean constraint list, writes the staff briefing, drafts parent notes about changed lessons, and proof-reads your rules for contradictions.
- A person: decides what the soft constraints are worth and signs off the result.
If you already run a smaller teaching business, the tutoring centre admin tutorial shows the same split on a scale with no separate timetabling office.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA small school worked through, with the sums
Take an illustrative school: eight classes, five days, six periods a day, so 30 lesson slots per class and 240 class-lessons a week. The staff room has 14 teachers, four of them part-time. There are two science labs, one computing room and a sports hall. These numbers are made up, but the sums are the ones you should run on your own data before touching any tool.
Sum 1: total demand against total supply. Teaching 240 class-lessons needs 240 teacher-periods. Ten full-time teachers at a 22-period teaching load supply 220. Four part-timers must therefore supply at least 20 between them. If their agreed loads are 6, 5, 5 and 4, that is 20 exactly, so there is no slack at all. Any sickness, any lost period for a trip, and the timetable cannot be completed without cover. A solver will report "no solution". A chat assistant will quietly produce a grid with a class missing a lesson in period six on Friday.
Sum 2: room capacity. Say science needs three lab-periods per class per week. Eight classes need 24 lab-periods. Two labs across 30 slots give 60 lab-periods of capacity, so labs are fine on paper. Computing needs two per class: 16 periods in one room with 30 slots. Fine, but if only the afternoon (periods 4 to 6) is allowed for computing because of when the technician works, the room has 15 usable slots and demand is 16. One lesson cannot be placed. That is a hard-constraint clash invisible until the sum is done.
Sum 3: part-time windows. A part-time teacher works Monday to Wednesday and has 6 lessons to teach. Available slots: 3 days × 6 periods = 18. Remove periods 1 and 6 on Monday because of a school-run request, and a lunch slot each day: 18 − 2 − 3 = 13 slots for 6 lessons. Comfortable. Now give that same teacher an exam-year practical taught as two doubles, each needing a pair of back-to-back slots that avoid the lunch gap, and the free choices narrow fast. This is the type of arithmetic that tells you where a solver will struggle before you press go.
Only after those three sums do you load the data. If a sum fails, you have found a staffing or room problem, not a software problem, and no tool can fix it.
Where part-time staff and room clashes slip through
Part-time staff are where most hand checks have to concentrate. A solver respects an "available" window, but the window rarely tells the whole story.
- Split sites or shared roles. A teacher who also works at another school may need a gap of an hour either side of travel. If you enter only "Tuesday and Wednesday", the solver can place a lesson at 08:50 and one at 15:30 with nothing between.
- Dual roles. A part-time teacher who is also the exams officer on Thursday has a day that looks free in the data and is not.
- Cover and duty. Break duty and detention rotas live outside most timetable inputs. Teachers can look compliant on paper and be double-booked in practice.
Room clashes take a different shape. Capacity and type are easy, so the solver gets those right. What escapes it is the shared room: the hall used for assembly on Monday, exams in summer, and outside hirers on Tuesday evening. Another is the "room that is technically free but not usable": a lab with no gas supply that week, a studio with a leaking roof. Neither appears in a data file unless someone adds it.
A before and after for a double-booked lab
Here is an illustrative mistake. A head of science pastes this into a chat assistant: "Eight classes, two labs, three science lessons each per week, spread across Monday to Friday, six periods a day. Output a timetable."
The assistant answers with a table. Period 3 on Wednesday lists Class 8C in Lab 1 and Class 9A in Lab 1. It has also given Class 10B science in period 6 on Friday and not noticed that the Friday afternoon was reserved for a trip. Nothing in the answer flags either problem, because the model had no way to test its own output.
The fix is not a better prompt. It is to send the same problem to a solver, then use the chat assistant afterwards to produce a "clash summary for the head of science": which rooms are busiest, which classes have two science lessons on one day, which days are lightest. That summary is the sort of task a language model handles well because you can check it against the grid.
Using a chat assistant for the paperwork around the grid
The unglamorous work of timetabling is gathering and tidying the rules. This is where a chat assistant saves real hours. Paste in the staff emails (with names replaced by codes) and ask it to extract the constraints.
Below are availability requests from teachers, with names replaced by codes (T1, T2...).
Turn them into a table with columns: teacher code, days available, periods to avoid,
hard or soft, source line. If a request is ambiguous or contradicts another, list it
separately under "Needs a decision" and do not guess.
[paste requests]
An illustrative reply gives a neat table and then a short "Needs a decision" list: for example, T4 says "Not Fridays" in one email and "Friday morning only" in another; T7 asks for no early starts but is listed as the only teacher of a subject with a period-1 double in the option block. Those two items are exactly what a person should settle before the solver runs, and spotting the contradictions is a strength of the language model.
Two more uses carry little risk. Ask it to draft the notice to parents when a lesson time changes, and to produce a one-page staff summary of what changed since last term. Both can be checked in minutes against the real timetable.
An illustrative parent note it might return: "From Monday [date], Class 8C's Wednesday science lesson moves from period 3 to period 5 so that both labs are free for practical work. Nothing else in the week changes. Please contact the office with any questions." What you fix: the reason (give one only if it is accurate and worth sharing), any class the change does not actually affect, and the date, which the model will guess if you leave it out. Paste the relevant rows of the real timetable into the prompt so the note is written from the grid, not from memory.
Manual review steps before anything is published
Whatever produced the timetable, run the same human check. It takes one to two hours for a small school and is the cheapest insurance you can buy.
- Run the tool's own clash report and keep a copy. It is your baseline. If the tool does not have one, count the lessons per class by subject and compare to the plan.
- Total check. For every class, add up lessons per subject and compare with the curriculum. A missing lesson is the commonest silent failure.
- Read each part-time teacher's week aloud. A human reading "Mon P2, P3, Tue P5" notices the problems that a rule set does not, such as no break between travel and teaching.
- Walk the room list. For each special room, scan every period it is used and ask whether it can actually be used then.
- Follow three pupils through a week. Pick one in the option blocks, one in a split group and one with a support timetable. If their week makes sense, most others will.
- Look at the worst day for each teacher. A timetable can pass every rule and still hand one teacher six lessons straight, plus duty. Fairness is your call.
- Ask two staff to find a problem. Offer a small prize. The people who live in the grid find the issue nobody wrote a rule for.
For schools that also run clubs or evening classes, the class timetable and booking data tutorial covers the demand-side question of when people want to attend, which hard rules do not capture.
What to ask a timetabling vendor claiming AI
If you are comparing paid products, a few specific questions cut through the marketing. Check the answers on the vendor's current pages or in a trial, not from a sales call alone.
- Does it handle part-time availability windows and "preferred but not required" slots as separate settings?
- Can it schedule double periods and keep a group in the same room for them?
- How does it deal with option blocks and pupils who cannot fit any pattern?
- Does it support two-week or rotating cycles if you run them?
- What does it do when no solution exists: tell you which constraints conflict, or just fail?
- Can you change one lesson mid-term and re-solve only the affected area, or does everything move?
- Does it export to your management information system or only to PDF?
- Where is pupil data stored, who is the processor, and can you run it without pupil names?
Anything else the vendor calls "AI" is worth a trial against a real term's data. In the guide to language school software with timetables, the same demo-with-your-own-data approach shows up quickly which features are real.
Pupil data stays out of chat tools
A timetable does not need names. It needs class codes, group sizes, subjects and teacher availability. Keep pupil records in your school system, and give any chat assistant anonymous codes and numbers. Special educational needs, safeguarding notes, medical information and anything about family circumstances should never be pasted into a general chat tool, even a paid one. The tutorial on putting customer data into ChatGPT explains what the plans do and do not promise, and the same logic applies to pupil data, which is more sensitive still.
If a vendor's solver is cloud-based, ask whether you can upload anonymised data and map codes back to names inside your own system afterwards. Many schools find that arrangement makes both the privacy and the review easier.
A term-planning routine that uses both
For a school of the size above, here is a sensible calendar, written as an illustration rather than a promise of time saved.
- Six weeks before term: collect staff and room requests; chat assistant turns them into a clean constraint list and flags contradictions.
- Five weeks before: run the three sums (demand, rooms, part-time windows) by hand or in a spreadsheet. Fix staffing gaps now.
- Four weeks before: load data into the solver; run; read the clash report; relax at most one soft constraint at a time.
- Three weeks before: the manual review above; send a draft to heads of department for comment.
- Two weeks before: freeze, publish, and use the chat assistant to draft the staff briefing and the parent note.
- Term time: log every change request. Repeated requests for the same fix are a rule you forgot to enter.
A final point on scale. For a school with a dozen classes, a spreadsheet and the three sums may beat any tool. For a school with fifty, the solver is the only sane option, and the human check matters more, not less. If you run a music or activity school rather than a full day school, the music school software tutorial will fit your size better.
What success looks like at the end of the first week
A timetable is not good because the software said it was. It is good when the first week passes without a surprise. Track these four numbers in week one: lessons where a teacher or room was missing, lessons run in the wrong room, staff requests for a swap, and pupils who turned up to the wrong place. Zero in the first two columns means the hard constraints were right. A pile of swap requests means the soft constraints and unwritten rules were wrong, and next term's constraint list just got better.
Keep that tally in a shared sheet and review it with the person who built the timetable. It turns timetabling from a once-a-year ordeal into a list that improves each term, which is where any tool, AI or not, earns its keep.
Questions heads of timetabling ask before trying one
Can ChatGPT or Claude make a whole school timetable?
For a handful of classes, a chat assistant can draft a grid, but it cannot prove the grid has no clashes. The bigger the school, the more likely a double-booking slips through, because nothing in the chat checks the grid. Use a timetabling solver for the grid and the chat assistant for the paperwork around it.
Is there free school timetabling software?
Yes. FET is free, open-source software for scheduling the timetable of a school, high school or university, released under the GNU Affero General Public License. Its own site says it usually solves a complicated timetable in at most 5 to 20 minutes, a simpler one in under five, and an extremely difficult one can take hours.
How long does a human check of a generated timetable take?
For a small school, plan one to two hours: run the clash check, read each part-time teacher's day, walk the room list, then look at three pupils' weeks. Skipping that hour is how a Monday-morning crisis starts.
Should I give the tool pupil names?
No. A timetable needs class codes, group sizes and teacher availability, not names. Keep pupil records in your school system and give the tool anonymous groups.
What if the solver says there is no solution?
That usually means two constraints cannot both be true, such as a part-time teacher who works two days but owns eleven lessons. Relax the softest constraint, or fix the staffing sum, and run it again.
Further reads
- AI Shift Scheduling vs a Rota Spreadsheet for Small Teams — Staff rotas and availability rules outside a classroom setting.
- AI Appointment Scheduling for Small Businesses: Tools and Setup — How slot-booking tools behave once real people book them.
- AI Use Case Template: Score Every Idea on One Page — Score the timetable idea against other jobs before you commit.
- What to Check in an AI Tool's Privacy Policy and Terms — What to read before any pupil-related data goes in.
- How Driving Schools Use AI to Keep Instructor Diaries Full — Another diary-filling problem with availability limits.
- How to Classify Business Data Before Using AI Tools — Sort school data into what may and may not leave your system.
- How Are Language Schools Using AI? Eight Admin Jobs and the Risks — Eight back-office jobs a language school can give to AI, from enquiry replies to homestay matching, with the specific risk and check for each.
- How Can AI Help School Administrators? Ten Jobs, With Prompts — Ten school office jobs AI handles well, each with a copyable prompt, a sample output and a check, plus the placeholder rule and where AI must not be used.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: FET project site (lalescu.ro/liviu/fet: licence and typical solving times, checked November 2026); timetabling vendors' home pages for how they describe AI features. Check each product's current features and pricing before buying.