Farming questions it can and cannot answer

The lower leaves have gone yellow. You took a photo, typed the question into an AI app, and got a clean answer back with a chemical name inside it. Before you spend money on that name, read this page. The tool was useful about some things and completely blind about others.

The three things it simply does not have

People ask an AI app three farming questions more than any others. Today’s weather. What my soil needs. Today’s mandi price.

It has none of the three.

It has no live weather feed. It has never seen your soil health card. It has no daily price list from your mandi. Ask anyway and it still answers, because answering is the habit it was built with. Any number it gives you there is a pattern, not a reading.

There is a quick test for this. If the honest answer begins with today, or this week, or this season, the AI is the wrong tool. A price, a forecast, a last date, a new rule. All of those changed after the model stopped learning.

Those three answers do exist. They just live somewhere else. Weather and crop advice for your district come from the government weather service, which issues district advisories twice a week, on Tuesday and Friday. The Meghdoot app carries them. Soil answers come from your own soil health card, made from a sample of your own field. Daily prices come from the government price portals, Agmarknet and eNAM.

Some apps can search the web while they answer. That helps a little, and how it searches, sees and uses tools explains what that really does. It still does not turn a chat app into a mandi board.

What it is genuinely good at

Four jobs. It can explain what a disease is, in plain words. It can explain what a line on your soil report means. It can explain a scheme document or a loan paper. And it can help you write down a clear question for an expert.

Notice what all four have in common. None of them need today’s data. They need meaning, and meaning is what this tool actually holds.

Here is a soil report prompt. Type your own card numbers in place of these.

I passed class 8. Explain in very simple English.
My soil health card says: pH 8.2, organic carbon 0.34 percent,
nitrogen low, phosphorus medium, potassium high.
What does each line mean in plain words?
Do not tell me how much fertiliser to use.
Then give me 5 questions to ask the agriculture officer about this card.

That last instruction matters. Researchers who tested general AI on crop advice for Indian farms found the same weakness every time. The advice comes out generic. The tool does not know your seed variety, your sowing date, or this season’s pest pressure in your block. It is also weaker on Indian material in general, which you saw in why it is weaker on Indian questions.

One habit makes every farming question work better. Write the whole thing in one language. Clean Hindi or clean English, not a mix of both. Tests keep finding that mixed writing makes the answer worse when facts matter.

If you speak your question instead of typing it, add one line at the end. Ask it to repeat back exactly what it understood. Village names, variety names and numbers get heard wrongly far more often than you expect.

Photograph it properly, then stop before you buy

A bad photo gives a confident wrong answer. Follow this every time.

• Shoot in daylight, not under a bulb. Morning or late afternoon light works best.
• Keep shadow off the leaf. Your own shadow is the usual spoiler, so stand to one side.
• Hold the phone steady and fill the frame with the affected part.
• Take three photos from three angles. Close on the spot, the whole leaf, then the whole plant.
• Put one healthy leaf beside the sick one in at least one photo.

Then type the things a photo cannot show. The crop and the variety if you know it. The sowing date and the age of the crop. The weather of the last week. Which part of the plant it started on. How many plants out of ten. Giving it your own notes and photos is the single biggest jump in answer quality, and this is that idea applied to a field.

Plantix is the strongest free crop app in India for this. Published tests put its accuracy above 90 percent. The same tests are honest about what moves that number. Light, blur, crop type and where the spot sits on the plant all matter. Rare diseases stay unreliable. A review of many plant disease apps found most were easy to use but weak at the real identification work.

One more thing. Take the photos before you touch or wash the leaf. People often clean the spot to make the picture look clearer. That wipes away the very thing an expert wants to see.

So the photo gives you a guess worth checking. It does not give you a result worth buying.

Two diseases can look the same on a leaf. A nutrient shortage can look like a disease. The app cannot tell you which, and neither can a chat.

There is another reason to slow down. The tool agrees with you very easily. Type that your neighbour says it is rust, and it will often build on that instead of testing it. So ask it flat. Give the symptoms, then ask for the three most likely causes and how to tell them apart in the field.

Even when the name is right, the quantity is a separate question. The dose, the water quantity, the spray timing and the safe waiting days before harvest are printed on the product label. That label matches the exact pack you bought and its strength. An AI has not seen your pack. Read the label, and have the officer or the shop’s qualified person confirm it against your crop.

From one yellow leaf to five good questions

Here is the whole method in one example. Wheat, sown in late November, lower leaves going yellow.

I am a farmer in Barabanki district, Uttar Pradesh.
Crop: wheat. Sown on 20 November. The crop is about 45 days old.
Problem: lower leaves are turning yellow from the tip inwards.
Upper leaves are still green. About 1 plant in 4 is affected.
It started 6 days ago, after heavy irrigation and 3 cloudy days.
Do not name any medicine.
Write my problem as one short paragraph I can read out on a phone call.
Then write 5 short questions in Hindi to ask the agriculture officer.

A good answer comes back in two parts. First a tidy paragraph you can read aloud without stumbling.

My wheat was sown on 20 November in Barabanki. The crop is
about 45 days old. The lower leaves are turning yellow from
the tip inwards. The top leaves are still green. About one
plant in four is affected. It began six days ago, after
heavy irrigation and three cloudy days.

Read that aloud once. It takes twenty seconds and it holds every fact the advisor needs. Now compare it with what most people say on that call. My crop is turning yellow, what should I spray. Nobody can answer the second version, however expert they are.

Then come questions close to these five.

1. Is this a nitrogen shortage or a disease?
2. What check can I do in the field to tell the difference?
3. If it is a shortage, what do I give, and how much, for my soil?
4. If it is a disease, what is its name and how does it spread?
5. What must I do this week to stop it reaching the rest of the field?

Now take that to a person. The block agriculture officer, the Krishi Vigyan Kendra in your district, or the government farmer helpline. The advisors on that helpline are agriculture graduates, and they answer in many Indian languages.

One caution about the number. Do not use a helpline number an AI gives you, because it may invent one. Find it on the agriculture department’s own website and check the web address ends in gov.in.

You have not replaced the expert here. You have walked in with a described problem instead of a vague one. That is the whole gain, and it is a real one.

Papers you can safely put in front of it

A scheme letter, a subsidy form, a loan paper from the bank. These are text, and text is where the tool is strong.

Here is the text of my loan paper. [type or paste the lines]
Explain each line in simple English, one line at a time.
List every charge and every date in a table.
Tell me which lines I should ask the bank to explain again.
Do not guess anything that is not in the text I gave you.

Remember the split. It knows the general rule, and only the office or the official portal knows your case. That split is the whole subject of understanding schemes, forms and offices. For PM-KISAN there is even an official government chatbot, called Kisan e-Mitra, reached from the scheme’s own portal and app. It works in several Indian languages and takes voice questions. It covers that one scheme and nothing else.

The quiet daily uses that actually pay

Planning and record keeping are safe, because nothing here depends on today’s price or today’s sky.

Ask it for a sowing and work calendar. Give it your crop, your state and your sowing date. Ask for a table with the week, the dates and the work due. Then ask it to mark the weeks where you must check with a local expert first.

Ask it to build an expense record. List your costs for one acre, item by item, and let it make the table and the total. Then ask for a simple profit line.

Here are my costs for 1 acre of wheat. [list them]
Make a table with item, amount in rupees, and date.
Add the total at the bottom.
Then show my profit if I sell 18 quintals.
Do not guess the selling price. Ask me for it.

That last line is the guard. The moment it invents a price, the whole sheet becomes fiction.

Keep that record going all season, in one place. A page in a notebook, or a note on your phone, is enough. At the end of the season, give the whole list back and ask for one summary table. Cost per acre, yield, and what you actually earned.

Most farms never get that last number written down anywhere. It is the number that tells you what to change next year.

Check the arithmetic yourself before you believe the total. It types numbers the way it types words, which is exactly why it gets sums wrong. A calculator and two minutes settle it.

What to take forward

The pattern here is simple. Facts about today come from the official source. Meaning, explanation and better questions come from the AI. Money decisions come after a person who knows your field has looked at it.

That same pattern works behind a counter as well as in a field. The next lesson, for a shop or a small business, applies it to prices, messages and records.