Crop Planning With AI: Rotations and Sowing Dates From Field Lists

Put each field on one spreadsheet row with its area, soil, access limits and the last three to five crops. Write your rotation rules as plain sentences, give the sowing windows you actually achieve, and ask AI to propose next season's crop per field and to show the machine and labour days each plan needs. Your agronomist then checks the draft.

A chat assistant is good at applying rules to 20 rows without getting tired, and at laying out the sums on drill days and combine days. It is bad at knowing your land. If you do not tell it that Clay Piece cannot be travelled before April, it will happily put spring beans there. So the work in this tutorial is mostly about the inputs: the field list, the rules, the windows, and the checks at the end.

Follow me on Instagram@sagnikteaches

The field list: what each row needs before AI sees it

The quality of the plan is capped by the quality of the list. Build one sheet with a row for each field and consistent columns. A usable layout:

Connect on LinkedInSagnik Bhattacharya

Spend the time on the crop names. Pick one list ("Winter wheat", "Spring barley", "Oilseed rape", "Spring beans", "Winter oats", "Spring oats") and use it everywhere. If your records are in a mess, using AI for farm records and paperwork covers how to get them into one place, and what a chat assistant can read in spreadsheets and photos explains what breaks when you upload a file that is half merged cells.

Subscribe on YouTube@sagnikteaches

Keeping private farm data out of the wrong tool

A field list is business data: land, yields, sometimes bank-linked arrangements. Use a business or paid plan whose terms say your content is not used for training by default, strip out anything you do not need to share (landlord names, account numbers), and keep the original file somewhere you control. The plan only needs field names, areas, soils and crops.

Writing rotation rules the AI can apply and you can audit

Rules turn "plan a rotation" from a guess into a check. Write them as numbered sentences. Take them from your own experience and your agronomist; the ones below are illustrative examples of the kind of rule, not agronomic advice, and the numbers must be yours.

1. No crop follows itself in the same field.
2. No crop from the same family follows another in the same field without a break crop between (cereal after cereal needs a break after two).
3. A break crop (a legume or oilseed) appears at least once every [N] years in each field. Set N with your agronomist.
4. Oilseed rape returns to a field no more often than once in [N] years.
5. Fields marked "late to travel" take winter-sown or late-spring crops only.
6. No more than [X] hectares of any one spring crop, so the drill and harvest are not squeezed.
7. Fields flagged with a weed or disease problem go on the "agronomist decides" list and are not planned by AI.

Rotation principles are well established: crops from the same family tend to share pests and diseases, so moving them lets those populations fall; different crops take different nutrients; dense crops can suppress weeds for the one that follows. The Royal Horticultural Society's rotation guidance puts this plainly for vegetables, and the same logic scales to field crops, though the families and the gaps differ. For specific disease gap years, use your agronomist's figures rather than anything a chat assistant recalls, because the intervals depend on the disease and on your soil.

A first plan: eight fields and 206 hectares, start to finish

An illustration. A mixed arable farm has eight fields totalling 206 hectares and wants a draft for the 2027 season. The owner pastes the sheet and the seven rules into a chat and uses the prompt below. Time to build the sheet from scratch the first year is around two hours; the second year it is a copy and update, nearer 30 minutes.

You are helping a farmer draft next season's crop plan. Use ONLY the
field list and rules I give you. Do not use general knowledge about
yields, prices or local conditions.

TASK 1: For each field propose the 2027 crop from this list: Winter
wheat, Spring barley, Oilseed rape, Spring beans, Winter oats, Spring
oats. Show a table: field | ha | last 3 crops | 2027 crop | which rules
applied | doubts. Mark any field where no option satisfies every rule
as "NEEDS HUMAN DECISION" instead of forcing a choice.

TASK 2: Total hectares per crop and check them against rule 6.

TASK 3: Do not propose varieties, seed rates, fertiliser or sprays.

RULES: [paste numbered rules]
FIELD LIST: [paste rows]

An illustrative result for the 2027 plan, after the owner's edits:

Hectares by crop: spring barley 61, winter wheat 62, spring beans 24, winter oats 22, spring oats 19, oilseed rape 18. Total 206. The owner checks that against the field list (24 + 31 + 18 + 27 + 22 + 35 + 19 + 30 = 206) before looking at anything else. If AI's total were 204 or 208, it has dropped or double-counted a field, and none of the later sums can be trusted.

Two rows carry a flag, and they are the most useful part of the draft. The AI marked Clay Piece and Low Meadow as NEEDS HUMAN DECISION: oats are a cereal, both fields have just grown two cereals, Clay Piece can only take an autumn-sown crop, and the obvious break for Low Meadow, spring beans, would take beans past the owner's rule 6 limit. The owner pencilled in oats so the workload sums could run, and put both fields at the top of the agronomist's list rather than letting the AI force a choice. Then the owner reads the "doubts" column, which is where AI says what it was unsure about. A useful draft says things like "Stable Field: second cereal in three years, check take-all risk with agronomist". If every row says "no doubts", rerun the prompt and ask it to challenge its own plan.

Sowing dates: windows come from you, placement comes from AI

A chat assistant cannot know your frost pattern, your soil warming or the variety you will buy. It can place dates inside windows you provide and check them against capacity. The window sources, best first:

1. Your own records: the dates you actually drilled each crop on each field over the last five years, including the late years.
2. Your seed merchant's or breeder's sowing guidance for the variety you plan to buy.
3. Your agronomist's advice for this year's conditions.

Put each crop's window and each field's earliest and latest realistic date in a second sheet. A field that is late to travel gets a later earliest date than the farm's general window. Then ask AI to place the plan:

Using the 2027 plan above and the window table below, propose a
sowing order for the spring crops. For each field give: start date,
end date, hectares, drill days needed (drill does [12] ha a day).
Assume only [35] percent of days in the window are workable. Show
the calculation. Flag any day where two fields compete for the
same drill, and say what I would have to change if the weather
only gives me half the workable days I have assumed.

Illustrative output: spring drilling covers 104 hectares (61 spring barley, 24 spring beans, 19 spring oats). At 12 hectares a day that is 8.7, so 9 drill days. The window runs from 1 March to 15 April, which is 46 days; at 35 percent workable, that is about 16 workable days. Seven days of spare capacity. If only 17.5 percent are workable (half as many), that is 8 days, and the plan fails by one day. The sensible reading is that beans go first (earliest and most sensitive to late sowing), barley second and oats last, with Low Meadow, the wettest field, held until the others are done.

The workable-days percentage is a judgement. Take it from your records ("in the last five springs I drilled on about X days between those dates") rather than letting AI pick. This is where a spreadsheet beats a polished answer: you can change the percentage and watch the plan break.

Labour and machinery limits that break a plan on paper first

A rotation that works agronomically can still be impossible to run. Add three limits to your sheet:

• Machine capacity per day for the drill, the sprayer, the combine and any contractor you rely on.
• People: how many operators, and any weeks they are away.
• Windows that overlap: harvesting early crops while drilling oilseed rape in August, for example.

An example of the sum that catches problems. With a combine doing 8 hectares a day, 206 hectares is 25.75, so about 26 combine days. If the harvest window is 45 days and half of them are workable, you have about 22 days. That is a shortfall of four days. The plan is not wrong, but it needs an answer: a contractor for four days, a second operator keeping the combine running longer hours, or a change in crop mix so harvest dates spread out. AI can test each: "What happens if I switch 20 hectares of spring barley to winter barley?" It can then recompute the days. Your rota matters here too, and the farm rota and timesheet automation tutorial shows how to turn those days into a team schedule.

Spreading the load by mixing winter and spring crops

Winter and spring crops ripen at different times, so the combine's work is spread. The risk is the opposite on drilling days: more winter crops means autumn is crowded, and more spring crops means spring is. Ask AI to show both peaks side by side for two or three crop mixes and compare "peak week hectares per machine". If one plan shows 70 hectares in a single week against a capacity of 60, that plan fails whatever its rotation score.

What your agronomist must check, and how to make that fast

The draft is a starting point; the sign-off belongs to the person who knows local conditions. To make the review take 30 minutes instead of two hours, give the agronomist a one-page pack, not a chat transcript:

1. The final plan table with each field's last three crops and the 2027 crop.
2. The "doubts" and "NEEDS HUMAN DECISION" rows highlighted.
3. The rules you gave the AI, numbered, so they can say which they disagree with.
4. The workload sums for drill and combine, with the assumptions listed.
5. A short list of your own questions: disease gaps, herbicide carry-over, varieties, nitrogen, and which fields need soil tests.

Their review covers what the AI could not know: disease and weed pressure by field, resistance, soil test results, input timing, and any scheme or contract conditions you farm under. If they change a field, update the sheet and rerun the totals; do not edit the table by hand without redoing the sums. For an example of how a similar write-up gets checked by an adviser, AI for crop-walk write-ups for an agronomist shows the other direction of the same relationship.

A plan that looked right until the spreadsheet was rechecked

An illustrative error. A farm manager asks AI to plan from a list in which two fields share a similar name, "Church Field" and "Church Field North". The assistant treats them as the same field and merges the histories, producing a plan that puts wheat after wheat on one of them. The totals come out 31 hectares short. Nothing in the answer looks odd except the total, and only because the manager adds up the hectares by hand.

Three habits prevent it. Give every field a unique name or code and use it in every column. Check the hectare total against your own number first. And ask the AI to list the fields it read before it plans: "List every field name you found and its hectares, count them, and stop." If the count is eight and you have nine fields, find out why before anything else.

Making the AI argue against its own plan

A first draft that satisfies every rule can still be a poor plan, because the rules are only as good as what you wrote. After the table is settled, ask for a challenge round in a fresh message:

Act as a sceptical farm manager reviewing this plan. Using only the
field list and rules above, give me the five weakest points: fields
where a rule is met only narrowly, years ahead where a field runs out
of valid crops, and any week where two jobs need the same machine.
For each, say what extra information would settle it.

An illustrative reply: "Stable Field meets rule 2 only because 2025 was spring beans; if that record is wrong, barley would be a third cereal in a row. Low Meadow can only take spring crops, and spring beans is the only spring break crop on your list; if rule 6 or the wet ground keeps beans off it, it has no valid break crop under rule 3. Clay Piece has no valid 2028 crop that follows winter oats and is also winter-sown, apart from oilseed rape, which rule 4 may block." The second point is a genuine structural finding: a field that can only take spring crops needs a spring break crop that suits it, or it will drift into a run of cereals. That is a decision for you and your agronomist, and it costs two minutes to discover now rather than in two seasons.

What-if runs that make the tool worth using

Once the base plan stands, a spreadsheet and an assistant can answer questions that used to take an evening with a pencil:

• "What if I drop oilseed rape?" Which fields lose their break crop, and what replaces it?
• "What if I rent 25 more hectares?" Drill and combine days with the extra land in each crop.
• "What if a wet spring halves workable days?" Which fields switch to a later-sown crop first, and what that does to the rotation rules.
• "What if the bean price falls?" Not for AI to answer from memory; give it your own margin figures per hectare and let it total the gross margin per plan, then check the sum yourself.

A simple gross margin comparison works with the figures you provide. If spring beans earn you $520 a hectare and spring barley $610 a hectare on your numbers, swapping the 24 hectares of beans to barley adds $2,160, but breaks the break-crop rule on Barn Field, which carries a cost in disease risk the sum does not show. Present every what-if as "agronomic cost" next to "financial gain", and let your agronomist weigh them. If you need these figures for a lender, farm business plan and lender records with AI covers how to present them.

Where a chat assistant stops and farm software starts

A sheet and a chat assistant handle a first draft and what-ifs for a farm of this size. They do not draw maps, hold field-level records through the season, or create the exports some schemes and buyers require. Dedicated farm management and satellite tools cover those jobs, with prices that vary by size; what satellite monitoring costs compares two. For a broader view of the choice, AI assistants for farmers: ChatGPT, Claude or a farm app sets the options side by side. A cautious order of work: plan in the sheet, check with the agronomist, then enter the agreed plan in whichever records system you use.

How to tell the plan is sound before you order seed

1. Totals reconcile. Hectares in the plan equal hectares in the field list, field by field.
2. Rules hold. Scan each row against each rule yourself, or ask AI to produce a pass/fail grid, and check two rows by hand.
3. Capacity holds. Drill and combine days fit the windows at the pessimistic workable-days figure, or you have a named fallback (contractor, extra operator, crop swap).
4. Doubts are cleared. Every "NEEDS HUMAN DECISION" row has an answer from you or the agronomist, written down.
5. Next year works too. Rerun the logic one year ahead; if the plan traps a field with no valid option in 2028, change 2027 now.
6. It is saved. The sheet, the rules and the prompt are in one folder with a date, so next year you start from this, not from nothing.

The plan is a decision aid; the field walk in March still decides whether Clay Piece is fit for drilling. If you would like to set up the sheet, the rules and the workload check together, that is what my AI implementation consultation covers.

Questions farmers ask about AI crop plans

Can ChatGPT or Claude plan my crop rotation?

Yes, as a first draft, if you give it a field list with crop history and your own rules. It applies rules quickly and shows its working. It does not know your soil, disease history or local conditions beyond what you supply, so your agronomist reviews the draft before any seed is ordered.

What should a field list contain for AI to use it?

One row per field: name, hectares or acres, soil type, drainage or access limits, the last three to five crops with years, known problems such as weeds or disease, and any constraints. The more consistent the column names and crop spellings, the fewer errors in the answer.

Can AI choose my sowing dates?

It can place sowing within windows you provide, such as your seed merchant's range and your own record of dates actually achieved. It should not invent windows from general knowledge, because suitable dates depend on soil temperature, frost risk and variety in your locality.

How do I check the AI's sums on workload?

Ask it to show each calculation: hectares by crop, drill or combine capacity per day, days needed, workable days available. Recalculate two crops by hand. If those match, the rest of the sheet is likely sound. Always recheck after any change to field areas.

Is dedicated crop planning software better than a chat assistant?

Dedicated software gives maps, records, and compliance exports, and suits larger or regulated operations. A chat assistant with a spreadsheet suits a first draft and what-if checks on a small or mid-sized farm. Many farms use both, with the draft checked in the software.

Further reads

• AI Chatbot for Farmers: Answers From Your Own Farm Notes and SOPs — Let a chatbot answer from your own farm notes.
• Free AI for Agriculture: What a Small Farm Gets Without Paying — What a small farm gets from free AI tools.
• Farm Accounting Software: QuickBooks, Xero or a Farm App With AI? — Where the crop plan's costs end up in the books.
• Best AI App for Farmers: Herdwatch, AgriWebb and Farmonaut Compared — Compare farm apps against a spreadsheet approach.
• Livestock Record Keeping App: Herdwatch vs AgriWebb AI Voice Notes — Voice-note record keeping if you also run livestock.
• AI Use Case Template: Score Every Idea on One Page — Score crop planning against other jobs AI could do on the farm.
• Flower Farm Business Plan: Bed Maps, Stem Pricing and Wholesale With AI — Plan a cut flower farm from stems per bed to price per bunch for markets, florists and subscriptions, with waste counted and AI doing the arithmetic.
• How AI Can Help Farmers Save Water: Build an Irrigation Log — A spreadsheet irrigation log fed by free weather data, with ChatGPT drafting the weekly schedule and you checking it against local crop advice.
• AI Chatbot for Farming: What It Can Answer and Where It Fails — What a chatbot can reliably do on a farm, with sample prompts, and the agronomy, livestock and safety questions it should never settle.
• AI Crop Monitoring on a Small Farm: Does It Work Without Drones? — How satellite-based crop monitoring works without drones, what it costs per hectare, and why fields under about a hectare need extra care.
• Do Farmers Use AI? Eight Uses of Artificial Intelligence, With Costs — How many farmers use AI, and eight small-farm uses with indicative costs, example prompts and the checks that keep each one honest.
• What Is Automation in Agriculture? Farm Jobs Ranked by Cost and Payback — Five small-farm jobs ranked by labour hours saved, with a cost and payback for each and a worked example on a mixed farm.
• Why Farmers Still Don't Use AI and What a Small Farm Can Try First — The four real barriers to AI on a farm, and a low-risk first week using paperwork, planning and supplier messages.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: RHS crop rotation guidance (rhs.org.uk, checked 5 November 2026) for rotation groups and the reasons to rotate; trade and extension material on break crops and rotation design; all field names, hectares, capacities and dates in the worked example are illustrative.