Why does AI give bad or made-up answers?

Steps

1. 1

   Give it the facts.

   Paste the email, the price list, the notes. If the answer depends on your data, the AI needs your data.

2. 2

   Tell it not to guess.

   Add "only use what I gave you" and "if you do not know, say so" to the prompt.

3. 3

   Ask it to show where the answer came from.

   "Quote the line from my notes that supports each point" makes made-up parts easy to spot.

4. 4

   Break big jobs into small ones.

   One task per prompt. Check it, then do the next.

5. 5

   Check anything that matters.

   Numbers, names, dates, laws, prices and links. Look them up yourself before you send or use them.

6. 6

   Start a new chat when it goes off track.

   Long chats can get muddled. A fresh chat with a clear prompt often fixes it.

Prompts you can copy

Answer only from my facts

Use only the information below to answer. Do not add facts from anywhere else.
If the answer is not in the information, say "not in the notes" instead of guessing.
Information: [paste notes, document or email].
Question: [your question].

Check your own answer

Here is an answer you gave me: [paste].
Go through it line by line. For each fact, number or name, tell me:
- Where it came from in what I gave you, or
- That you were not sure or filled it in.
Then give me a corrected version with only the facts you can back up.

Ask me first

I need [task]. Before you write anything, ask me up to 5 short questions about facts you would otherwise have to guess.
Wait for my answers. Then write it using only my answers and the facts I gave.

Example

This is a made-up example. You ask: "Write a short note to a customer about our return policy."

The AI does not know your policy, so it writes "returns are accepted within 30 days with a receipt." That sounds right, but your real policy is 14 days and no returns on special orders. If you send it, you just promised something you do not offer.

Now try: "Here is our return policy: [paste]. Write a short note to a customer explaining it. Only use what is in the policy."

The answer now says 14 days and mentions special orders. Same tool. The difference is that it had the facts.

Tips

• Bad answers are often a prompt problem. If the AI keeps missing, add facts before you add more rules.
• Watch for answers that sound very sure about very specific things, like an exact price or a law. That is where made-up answers hide.
• Some tools can search the web or read your files on some plans. That helps, but still check the sources it gives you.
• Ask the same question two ways. If the answers disagree, check both.

What not to do

• Do not use AI for a final answer on legal, tax, medical or safety questions. Use it to prepare questions for the real expert.
• Do not trust a link, quote or study an AI gives you until you open it and read it yourself.
• Do not keep arguing in a long chat that has gone wrong. Start over with a better prompt.

For better prompts from the start, read how to write a good prompt or grab one from the prompt library.

FAQ

What is an AI hallucination?

It is the name for when AI states something false as if it were true, like a fake quote or a wrong date. It happens because the AI predicts likely words, not checked facts.

Is one AI tool better at not making things up?

The tools keep changing and each has good and bad days. Any of them will guess when facts are missing, so the fix is the same: give it facts and check the answer.

Can I trust AI math?

For simple math it is usually fine, but check anything that matters. For totals and prices, use a calculator or your spreadsheet.

Why is the answer different when I ask again?

AI tools do not give the same words every time. If the facts change between answers, that is a sign it is guessing, so give it the facts.

Rather have it done for you? We build the system that handles your outreach, one time, and you own it. See how it works.

First posted October 28, 2026. Last updated October 28, 2026. Free to copy and use, CC BY 4.0.