BeginnerLesson 9 of 305 min read

Common mistakes and good habits: accuracy, privacy and honest use

Learn the ten most common prompt mistakes, the real limits of AI answers, a safe-use routine for different stakes, and privacy and honesty habits.

In this lesson you will

  • Recognise ten common prompt mistakes
  • Know the limits of AI answers
  • Match checking to the stakes
  • Protect privacy and use AI honestly

About 16 minutes

What goes wrong, and how to see it coming

Kamran asks an AI for statistics about online learning for a school presentation and pastes the numbers straight onto a slide. A teacher asks where they come from. He cannot say, and the numbers cannot be found anywhere. The AI produced figures that looked right. Kamran learned the most important habit in this course the hard way: check anything that matters. This lesson covers the common mistakes, the limits of AI answers, and the habits that keep you safe and effective.

Ten common prompt mistakes

Mistake What happens Fix
Vague task Generic answer Use a clear verb and finish line
No audience Wrong tone or level Name who will read it
Missing facts Invented details Supply facts, say use only these
No format Too long or messy Specify length and structure
Too many tasks Some parts rushed One job per prompt, or number them
Conflicting rules Random choice between them Remove contradictions
Pasting everything Focus is lost Say which part matters
Giving up after one try Poor results Iterate with one change at a time
Trusting without checking Errors reach your work Verify important claims
Sharing private data Privacy risk Remove or anonymise before pasting

Limits you must know

  • Hallucination. Models can state false things fluently, including invented numbers, quotes, links and sources.
  • Old or missing knowledge. A model may not know recent events, prices or rules, unless the tool searches the web.
  • No lived experience. It cannot see your local situation or confirm facts about people.
  • Pleasing bias. Models tend to agree with you. A leading question may get a flattering answer.
  • Inconsistency. The same prompt may give different answers each time.

A safe-use routine

  1. Decide the stakes. Low (a party invitation), medium (a school essay), high (health, money, law, safety).
  2. Match the checking. Low: read it. Medium: verify key facts from a trusted source. High: use a professional.
  3. Ask for uncertainty. Add: ‘If you are not sure, say so and say what I should check.’
  4. Ask for sources, then open them. A source that does not exist is a warning sign.
  5. Keep a human decision. You decide and you are responsible for what you send or submit.
Answer the question below. Mark any claim you are not certain about with [CHECK]. At the end, list what I should verify with a reliable source. Do not invent sources.

Question: [your question]

Privacy habits

  • Do not paste passwords, ID or card numbers, medical records or other people’s private details.
  • Replace names with labels such as ‘Customer A’ when you only need the pattern.
  • Check the tool’s settings on whether chats are used to improve the service, and change them if you prefer.
  • At work or school, follow the rules about which tools you may use.

Fair and honest use

  • If a teacher or employer has rules about AI help, follow them and disclose when required.
  • Do not present AI-written text as your own where that is not allowed.
  • Do not use AI to deceive, to harass, or to make fake reviews or fake news.
  • Use AI to learn, not only to finish. Try the task yourself first, then compare.

Your pre-flight checklist

A checklist turns good intentions into a habit. The small program below prints a score for a prompt against five questions. It is a manual checklist in code form, with the ticks set by hand. The output is real:

checklist = [
    ("States a clear task verb", True),
    ("Gives the audience", True),
    ("Supplies the facts to use", False),
    ("Sets a length or format", True),
    ("Says what to do if information is missing", False),
]
passed = sum(ok for _, ok in checklist)
print(f"Score: {passed}/{len(checklist)}")
for item, ok in checklist:
    print(("[x] " if ok else "[ ] ") + item)
Score: 3/5
[x] States a clear task verb
[x] Gives the audience
[ ] Supplies the facts to use
[x] Sets a length or format
[ ] Says what to do if information is missing

Here the prompt is missing the facts and the “what to do if information is missing” rule, so it is at risk of invented details. The fix is two short lines added to the prompt.

Practice: spot the problems

Read this prompt and list at least four problems before you read the answer: Write something good about climate change for my teacher soon and make it sound smart and include some statistics.

Practice answer

Problems: vague task (‘something’), fuzzy words (‘good’, ‘smart’, ‘soon’), no audience or length, no format, and ‘include some statistics’ invites invented numbers. A better prompt gives the topic, grade level, length, and says to use only statistics from a source pasted below, or to mark them [CHECK].

Common mistake

Using AI output as evidence for another AI prompt. If an AI invented a number and you feed it back as ‘a fact’, the error becomes harder to see. Check facts at the source.

Check what you learned

Five quick questions. Pick an answer and you will see straight away whether it is right and why.

What is a hallucination?

  1. A spelling error
  2. A slow reply
  3. A logo
  4. A fluent but false statement by the AI

Invented facts can look convincing.

Which stakes need a professional?

  1. A party invitation
  2. High stakes like health, money, law and safety
  3. A shopping list
  4. A joke

Use real experts for important decisions.

What does asking for uncertainty do?

  1. Makes the AI quieter
  2. Deletes facts
  3. Helps reveal what to verify
  4. Adds images

It prompts honest flags.

What should you do with sources the AI gives?

  1. Open and check them
  2. Assume they exist
  3. Ignore them
  4. Copy them blindly

Invented sources are a warning sign.

Which is a good privacy habit?

  1. Paste ID numbers
  2. Share passwords
  3. Replace names with labels like Customer A
  4. Post chats publicly

Anonymise before pasting.

Summary

  • Most failures come from vague tasks, missing facts and no checking.
  • AI can state false things fluently.
  • Match your checking to the stakes.
  • Protect private data and follow the rules for honest use.

Next lesson

Next: the beginner project, build your own starter prompt pack.

Questions

How can I tell if an AI answer is wrong?

Check key facts against a trusted source, look for invented names, numbers and links, and ask the AI what it is unsure about.

Is it cheating to use AI for school?

It depends on your school's rules. Follow them, disclose when required, and use AI to learn rather than only to produce answers.

Can I trust AI for health or legal questions?

Use it for general understanding only. Decisions need a qualified professional.

Progress is saved in this browser.