Lesson 5 of 6 · 5 min read · Beginner

Basic limitations and when not to trust it

AI can be fluent and wrong at the same time. Learn the three signs of a made-up answer, which kinds of question need checking, and practise telling a trustworthy reply from a confident one.

In this lesson you will

Sort tasks into trust, spot-check and verify
Spot the signs that an answer may be invented
Know what to do with a claim you cannot check

Lesson 1 said the model predicts what comes next rather than looking things up. This lesson is about what that means in practice. It is not a reason to avoid AI. It is a reason to know which of your questions it can answer from what you gave it, and which it might be answering from thin air.

Trust depends on where the answer comes from, not on how good it sounds.
Kind of question	Where the answer comes from	What to do
Rewrite, summarise or explain text I pasted	Your text	Trust it; read it once
Plan, list, structure or brainstorm	Common patterns	Trust it; adjust to taste
A specific fact, number, date, price or quote	Memory of training text, which may be fuzzy	Verify against a real source
Anything legal, medical, tax or financial for your situation	General patterns, not your circumstances	Use as a starting point, then check with a person or an official page
Something recent, like this month's prices or news	Nothing, unless it searched the web	Ask whether it searched, and click the source

Three signs an answer may be made up

A very specific detail you did not give it. A name, a page number, a statute, an exact figure. Specificity feels like evidence but is easy to generate.
A source you cannot open. A book title, a study or a web page that sounds right. Check that it exists before you repeat it.
Confidence with no hedging on a question that deserves some. Real experts say "it depends" and "check the current page". So do the better AI answers.

Which would you trust?

You ask how much a ChatGPT Plus subscription costs right now. Which answer should you trust more?

Lesson FAQ

Why does AI make things up?

Because it produces the most plausible continuation of your text. If the most plausible-sounding answer is a specific figure, it may produce one even when it does not know. It is not lying; it has no separate sense of true and false to check against.

Do more expensive AI plans make fewer mistakes?

Stronger models make somewhat fewer, but no plan removes the problem. Checking facts against a real source matters far more than which tier you are on.

Is it safe to ask AI about legal or medical questions?

Safe to ask as a starting point, to understand terms or prepare questions. Not safe to act on alone. Your situation has details it does not know, and a wrong answer in these areas is expensive.

How do I know if the AI searched the web?

Tools that search usually show source links under the answer. If there are no links and you asked about something recent, assume it did not search and ask it directly, or use a tool built for research.

Finished reading?

Mark it done to track your progress through the course.

Save progress across devices

← PreviousOne useful task you finish today Next →Where to go next: which course, which tool, and do you need to pay

Compare, calculate, decide

Which AI Should I Use? →How to Talk to AI →Best AI under $20 a month →ChatGPT vs Claude →Decision engine →

Three signs an answer may be made up

Lesson FAQ

Stop overpaying for AI tools