ChatGPT Prompts for Farm Admin: A Ready-to-Use Library

Most farm office work runs on the same few jobs: turning scribbled notes into records an inspector will accept, making sense of letters from the RPA, chasing money and checking sums. ChatGPT handles all of these well if you give it a clear brief. If you just type “help me with my SFI”, you’ll get vague, generic advice.

This page is a library of prompts you can copy, fill in and use today. Each one has been written for UK holdings, and the worked examples use real figures so you can see what good output looks like. If you want the wider picture first (which tools exist, what they cost across a year, and where AI fits on a mixed farm), start with our overview of ChatGPT and AI for farm admin.

How the prompts in this library are built

Every prompt below follows the same four-part pattern, because it consistently gets better answers:

1. Role: tell ChatGPT who it is acting as (“You are an experienced UK farm secretary”).
2. Context: the facts of your farm and the job in hand.
3. Task: exactly what you want done.
4. Format: how you want the answer laid out, such as a table, a letter or a checklist.

Anything in [square brackets] is a gap for you to fill in. Delete the brackets once you’ve done it. If a section doesn’t apply, delete the whole line rather than leaving it blank, as blanks tend to get filled with guesses.

One line belongs in almost every prompt: “If you are unsure of any rule, date or figure, say so rather than guessing.” It doesn’t stop mistakes altogether, but it noticeably cuts down on confident nonsense about scheme rules.

What to keep out of the chat box

Before the prompts, a rule. In ChatGPT, go to Settings → Data Controls and turn off “Improve the model for everyone”. On the free and Plus plans, your conversations can be used for training unless you switch this off.

Even with it off, don’t paste in:

• Bank details, sort codes or your Government Gateway login
• Your SBI number or CPH number when they aren’t needed for the task
• Neighbours’ or employees’ personal details
• Full scanned documents that carry all of the above

You can swap in placeholders like “CPH: XX/XXX/XXXX” and put the real numbers back once you’ve copied the text into your own records. ChatGPT doesn’t need your holding number to tidy up a medicine book.

Paperwork and compliance prompts

Decoding an RPA or Defra letter

You are a UK farm business adviser who explains government letters in plain English.
Below is a letter I have received from the Rural Payments Agency.
Tell me: (1) what it is asking me to do, (2) the deadline, (3) what happens if I miss it,
(4) any documents I will need. Use short bullet points.
If anything in the letter is ambiguous, list it as a question I should ask the RPA.

[paste letter text, with SBI and bank details removed]

This works well on payment statements, inspection notices and requests for evidence. Check the deadline it gives against the letter itself, because it does sometimes read the date of issue as the due date.

Building an SFI evidence diary

You are helping a farmer keep evidence for Sustainable Farming Incentive actions.
My agreement includes these actions: [list action codes and names, e.g. CSAM1, SAM1].
For each action, produce a table with columns: Action | What evidence to keep |
When to record it | Example diary entry. Keep it specific to a [arable / mixed / livestock]
farm in [county]. Flag anything I should confirm against the current SFI handbook.

SFI rules have changed several times since 2023, so treat the “what evidence to keep” column as a draft checklist. Check it against the scheme guidance on GOV.UK before relying on it.

Red Tractor pre-audit checklist

I have a Red Tractor [Dairy / Beef and Lamb / Combinable Crops] assessment in [number] weeks.
Create a pre-audit checklist grouped by: records, medicines, storage, animal welfare / crop
protection, and staff training. For each item, give a one-line "how to check it" note.
Mark the items most commonly picked up as non-conformances.

Livestock record prompts

Medicine book from vet notes: a full worked example

This is where ChatGPT probably saves the most time on livestock units. Say the vet has been out and you’ve jotted this on your phone:

3/10 vet Jones. 6 heifers coughing, pen 4. tags 0412 0415 0419 0423 0430 0431.
metacam 20mg/ml 2.5ml/100kg sub cut, all approx 350kg.
batch 4H7721 exp 06/27. withdrawal per label meat 15 days. given by me (Tom).

The prompt:

You are a UK farm records assistant. Turn my notes below into a veterinary medicine
record suitable for a Red Tractor assessment. One row per animal. Columns:
Date | Animal ID | Product | Batch | Expiry | Dose (ml) | Route | Meat withdrawal ends |
Administered by | Reason. Calculate the dose from weight and the withdrawal end date
from the treatment date. Show your dose calculation once beneath the table.
Use only the withdrawal period I give you; do not look one up.

[paste notes]

What came back:

Dose calculation: 350 kg ÷ 100 × 2.5 ml = 8.75 ml per animal.
Total used: 6 × 8.75 ml = 52.5 ml.

Check two things every time: the arithmetic (350 ÷ 100 × 2.5 is indeed 8.75) and the date count (3 October plus 15 days is 18 October). The line “Use only the withdrawal period I give you” matters. Without it, ChatGPT will sometimes quote a withdrawal period from memory, and the figure on the product’s data sheet is the only one that counts.

The finished table pastes straight into Excel, Google Sheets, or the import screen of herd software such as Herdwatch or Uniform-Agri.

Checking a cattle movement batch

Below is a list of cattle I am moving off farm on [date], copied from my herd records.
Check for: duplicate ear tags, tag numbers in the wrong format for UK cattle,
animals that appear to be under a medicine withdrawal period based on the treatment
dates listed, and any gaps in the data. Do not submit anything; just report problems.

[paste list: tag, DOB, sex, breed, last treatment date, withdrawal end]

You still report the move on the Cattle Tracing System (or BCMS’s online service) yourself. The prompt is just a second pair of eyes before you do.

Cropping and field record prompts

Spray records from a day’s notes

You are an arable records assistant. Convert my notes into a pesticide application record
with columns: Date | Field name | Area (ha) | Crop & growth stage | Product | MAPP number |
Rate per ha | Total product used | Water volume (l/ha) | Wind speed & direction | Operator.
Calculate total product used. If any column is missing from my notes, leave it blank
and list the missing items at the end.

[paste notes]

That last instruction stops ChatGPT inventing a MAPP number, which it will otherwise do quite happily. If you use Gatekeeper or Muddy Boots, ask for the columns in the order your import template expects.

NVZ livestock manure loading

The 170 kg N/ha whole-farm limit for livestock manure is a sum most farmers can do in their heads, but the paperwork wants it shown. Here’s a worked example for a 180-cow herd on 120 ha:

I farm in an NVZ in England. Calculate my whole-farm livestock manure nitrogen loading.
Stock: 180 dairy cows (6,000 to 9,000 litre yield), using a standard figure of 101 kg N
per cow per year. Farmed area: 120 ha. Show each step. Compare the result to the
170 kg N/ha limit and state the headroom in kg N and in cow numbers.

Expected answer:

Total N: 180 × 101 = 18,180 kg N/year
Loading: 18,180 ÷ 120 = 151.5 kg N/ha
Headroom: (170 × 120) − 18,180 = 20,400 − 18,180 = 2,220 kg N
Equivalent to: 2,220 ÷ 101 ≈ 21 further cows at the same yield band

Always give ChatGPT the standard N figure from the official table yourself. If you ask it to supply the figure, you’re trusting its memory of a government table, and that memory isn’t reliable. Followers, youngstock and imported manure all need adding in the same way on a real holding.

Money and office prompts

Contractor invoice

Write a contractor invoice from [business name] to [customer] for:
drilling 42 ha winter wheat at £55/ha, completed [date], field names [list].
Add VAT at 20%. Payment terms 30 days. Include a line for my VAT number: [VAT no.].
Lay it out as a plain table I can paste into Word.

The totals should come back as £2,310.00 net, £462.00 VAT and £2,772.00 gross. If they don’t, the prompt has a typo. The CAAV and NAAC contractor surveys are a better guide to rates than anything ChatGPT says, so set the rate yourself.

Polite payment chaser

Write a firm but friendly email to [customer] chasing invoice [number] for £[amount],
now [number] days overdue. We have a good long-term relationship. Under 120 words.
Offer to talk by phone. No legal threats in this first reminder.

Sorting a bank export before the accountant sees it

Below is a CSV export of my farm bank account for [month]. Categorise each line into:
feed, fertiliser, seed, sprays, vet & med, fuel, contractors, machinery repairs,
milk/livestock/grain sales, subsidy & scheme payments, drawings, other.
Return a table with the original line plus a Category column, then a summary of totals
by category. Flag anything you could not categorise confidently.

Remove account numbers before pasting. The categorised file goes to your accountant or into your Making Tax Digital software (Xero, QuickBooks, Sage or Farmplan’s Cashbook). ChatGPT doesn’t file anything with HMRC.

Where these prompts go wrong

Three problems account for nearly every bad result we’ve seen:

• Out-of-date rules. Scheme details, closed periods and payment rates change. ChatGPT’s knowledge has a cut-off date, and it won’t always tell you that. Anything regulatory needs checking on GOV.UK or with your agent.
• Arithmetic in long tables. Totals over 30 or so rows sometimes drift. Asking it to “show the calculation” makes errors easier to spot, and a quick SUM in a spreadsheet is the final check.
• Filling gaps. Give it half a record and it’ll complete the rest plausibly. This is why so many prompts here say “leave it blank and list what’s missing”.

Which tool, and what it costs

All the prompts above work in any of the main assistants. The differences are mostly price and where your data goes.

Prices move, so check the provider’s page before you subscribe. For most single-farm offices, Plus pays for itself if it saves about an hour a month.

Keeping your library somewhere you’ll actually use it

The quickest setup is a single note on your phone or a Word document called “Farm prompts”, with each prompt under a clear heading and your usual facts already filled in (herd size, hectares, county, scheme agreement actions). On ChatGPT Plus, a Project called “Farm office” lets you write those standing facts once in the project instructions, so each new chat starts already knowing you run 180 cows on 120 ha in Shropshire.

When a prompt gives you something you’ve had to correct, edit the saved prompt there and then. After a season of this, your library will reflect how your farm records things.