Lesson 26 of 64 Module 4: When it is wrong

Your checking habit on one page

8 min read Free, no sign-up 9 August 2026

After this lesson you can

  • Apply one short checklist to every answer that matters.
  • Decide in five seconds whether an answer needs checking.
  • Recognise your own weak spots.
  • Teach the checklist to someone else.

Read first: The two-chat test and three other checks , Checking sources, links and quotes

You asked the AI something. The answer came back fast, clean and sure of itself. Now you have to decide whether to act on it, and nobody near you can tell you if it is right.

This lesson is the whole of module 4 on one page. Six questions. About a minute. No new ideas here, just the habit.

The whole thing in six lines

Here is the checklist. Copy this block exactly as it is.

1. Is it rare, local or time-bound?
2. Did I give it the source, or ask from memory?
3. Did I ask twice, in two fresh chats?
4. Did I open every link?
5. Did I do the maths myself?
6. Does an official source agree?

Line 1 sorts the risk. A rare fact, a local fact or a date is where the model guesses most. Time-bound means it changes with the date. A last date, a fee, a price, this year’s syllabus.

Line 2 is the single biggest fix. Paste the notice, then ask about the notice. That is far safer than asking from memory. You saw this in giving it your own book, notes or photo.

Line 3 is the cheapest test there is. Ask the same thing in two new chats. If the two answers do not match, neither one is knowledge. You met it as the two-chat test in the two-chat test and three other checks.

Line 4 is the one people skip. A link that opens is not the same as a link that says what the AI claimed. You practised this in checking sources, links and quotes.

Line 5 is for anything with rupees, marks or measurements in it. Do the final sum on your phone calculator yourself. Line 6 covers only four things: money, medicine, law and government rules. There the AI is where you start. The official site or the office is where you finish.

Five seconds to decide if it is worth checking

You cannot check everything. If you try, you will stop using the tool inside a week. So one faster question comes before the six.

Ask yourself: if this is wrong, what does it cost me?

Check when being wrong costs money, marks, medicine or a journey. Money means a fee, a loan, a subsidy amount or anything you buy. Marks means anything going into an exam answer or an assignment. Medicine means any dose, any symptom, any treatment. A journey means a trip to an office, a bank or a college that you would make because of this answer.

If none of those four apply, do not bother. Ideas for a speech, a rewritten message, an explanation you will test yourself on anyway. Use the answer and get on with your day.

Running it on an answer that was wrong

Your cousin says there is a scholarship for class 12 students in your state. You ask the AI about it. The reply comes in four seconds and looks like this.

The last date to apply is 31 October.
The scholarship gives Rs 12,000 per year.
Apply on the state scholarship portal.

None of that is real. It is what a confident wrong answer looks like. Now run the six lines.

Line 1. Rare, local and time-bound, all three. A state scheme, an amount and a date. That is the highest risk there is.

Line 2. You gave it no source. You asked from memory. Second warning.

Line 3. You open a fresh chat and type the same question. This time the last date is three weeks different. Two answers, both calm, both exact. Neither one is knowledge.

Line 4. You open the link. It goes to a department home page. Nothing on it mentions that scheme or that amount.

Line 5. No maths in this answer. Skip.

Line 6. This is money and a government rule, so the official source decides.

Line 3 caught it in about forty seconds. Without that check you would have planned your month around a date that does not exist.

So where is the real date? Search for the department’s own website and check the address ends in gov.in. Or walk into the block office and ask. That method still works next year. Understanding schemes, forms and offices shows you how to split the work between the two.

Careful Never act on an AI answer alone for money, medicine, law or a government rule. Confirm it on the department's own website, or at the office. In India, trained lawyers have been fined for filing court papers with case references that an AI invented.

Running it on an answer that was right

Checking is not the same as distrusting. When you paste the numbers in yourself, the answer is far more likely to hold. You find that out in about a minute.

Your father has a loan paper. It says Rs 20,000, for one year, at simple interest of 12 percent per year. You want the total to repay. You type the numbers straight off the paper.

Here are the exact words from a loan paper.
Rs 20,000, one year, simple interest at 12 percent per year.
Work out the interest and the total amount to repay.
Show every step.

The answer says interest is Rs 2,400 and the total is Rs 22,400. Now run the six lines.

Line 1. Not rare, not local, not time-bound. Simple interest is ordinary school maths, written about everywhere.

Line 2. You gave it the source. The numbers came off the paper, not out of its memory.

Line 3. A fresh chat gives the same two numbers.

Line 4. No links, and none needed.

Line 5. You do it yourself. 20000 times 12 is 240000. Divide by 100 and you get 2400. Add 20000 and you get 22400. It matches.

Line 6. The rate came off the bank’s own paper, so the sum holds. The total you will actually repay is a different question. A loan paper can carry a fee, an insurance charge or a rate that changes later, and none of that was in what you pasted. Ask the bank for the full repayment figure in writing before you sign.

One minute, and the arithmetic stood. That is what a passed check looks like. Notice that line 6 never switches off for money. Why it gets sums wrong explains why line 5 still matters when the method looks right.

Two checks that do nothing

Two habits feel like checking and are not. Both need naming, because they creep back.

The first is typing are you sure. The model may say yes, more firmly than before. Or it may drop a correct answer because you pushed. Neither tells you anything about the world. A changed answer is a reason to go and check a real source. It is not evidence in either direction. It agrees with you far too easily shows how far this goes.

The second is being satisfied because the answer sounded confident. These tools almost never say they do not know. A calm, fast, very exact answer is also the normal shape of a wrong answer. An exact detail is often a sign of guessing, not of knowing. Compare in autumn with on 30 September. The second sounds more expert and is more likely to be invented.

Which of these five need checking?

Read the five. Decide check or do not check for each one. Answers are below, so cover them with your thumb.

  1. The AI explains how to work out a percentage and gives an example sum.
  2. The AI says the last date to apply for a state exam form is a certain date.
  3. The AI gives you ten ideas for a speech at a school function.
  4. The AI names the nearest ITI to your village and gives its address.
  5. The AI tells you how much fever medicine to give a child.

Now the answers.

  1. Do not check. Percentages follow a pattern, and it is written about everywhere. Test it on a sum whose answer you already know.

  2. Check. A date is time-bound and the form is official. Line 6 decides this one.

  3. Do not check. Nothing costs you anything if one idea is weak.

  4. Check. A named place near your village is rare and local. That is exactly where names and addresses get invented. Go and look, or ask someone who has been there.

  5. Do not use the AI at all. Go to a doctor or a chemist. This is not a checking problem.

If you marked 2, 4 and 5, the triage rule is already working in your head.

Copy it into your notes app before you close this page

Do it now, not later. Open your notes app. Make a note called My AI check. Type the six lines in.

Then use it. Not on everything. On the answers where being wrong costs you something real.

After three or four weeks you will notice which line keeps catching things for you. It may be line 3, or it may be line 4. That line is your weak spot, and knowing it is worth more than reading this lesson twice.

This one habit is the whole difference. The person who uses AI well is not the one who knows more tricks. It is the one who runs six questions before acting. The rest find out in front of an officer, a teacher or a shopkeeper.

Module 5 comes next, and it is about learning with AI. It opens with the answer machine trap. There the danger is not a wrong answer at all. It is a right answer that you did not work for.

Do this now

Run the six lines three times this week

  1. Open your phone's notes app. Make a new note and call it My AI check.
  2. Type the six lines from this lesson into it by hand. Typing helps you remember them.
  3. Ask the AI three questions you actually need answers to this week.
  4. Run the six lines on each answer. Give it about a minute each time.
  5. Write beside each answer which line caught something, or write nothing caught it.
  6. After three answers, look at your notes. The line that keeps catching things is your weak spot.

Remember this much

  • Six questions, about one minute. That is the whole habit.
  • Check when being wrong costs money, marks, medicine or a journey.
  • Two answers that do not match mean neither one is knowledge.
  • A link you did not open is not a source.
  • Are you sure is not a check. It only tests how hard you pushed.
  • For money, medicine, law and government rules, the official source decides.

Questions people ask

How do I check if an AI answer is true?

Ask the same question again in a fresh chat. If the two answers do not match, neither one is knowledge. Then open every link it gave you and see whether that page really says what the AI said. For money, medicine, law or a government rule, finish on the official website or at the office.

Do I need to check every AI answer?

No. Check when being wrong costs you money, marks, medicine or a journey. For an essay idea or a rewritten message, use the answer and move on. Trying to check everything is how people give up on the habit completely.

Does asking are you sure make the AI more accurate?

No. It often agrees with you harder, or flips to a new answer because you pushed. A changed answer is not proof that the new one is right. Treat any change as a signal to go and check a real source.

Can I trust AI for government scheme details?

Use it to understand what a scheme is and what the words on the form mean. Do not use it for the last date, the amount, the fee or the eligibility rules. Those change, and the AI is weakest exactly there. Confirm them on the department's own website or at your block office.

How long does the checklist take?

About a minute for most answers. Asking again in a second chat is the slowest part, and that takes well under a minute on a phone. The official source check happens later, on the department website or at the office.

Prices, free limits and app screens change often. The facts in this lesson were checked on 9 August 2026. If what you see on your phone looks different, trust your phone and read the idea, not the exact button name.

See all 64 lessons