The first answer is a draft. Learn the four follow-ups that fix most of them, ask the model what it assumed, and practise telling a trustworthy answer from a confident one.
In this lesson you will
→Match a weak answer to the follow-up that fixes it
→Use "what did you assume" to surface silent guesses
→Recognise an answer that is confident but not trustworthy
People who get good results from AI are not writing perfect prompts. They are asking one more time, well. The model remembers the whole conversation, so a follow-up is cheaper than a rewrite and usually better, because you now know exactly what was wrong.
Four kinds of weak answer and the follow-up for each. The pattern is always: name the problem, protect what worked.
The answer was
Say this
Why it works
Generic
"Too general. I am [situation]. Redo it for that."
Adds the context you skipped
Wrong shape or length
"Same content, as five bullets under 80 words."
Fixes output without touching the substance
Wrong in one detail
"The date should be the 14th, not the 12th. Change only that."
Stops the model rewriting everything to fix one thing
Confident but suspicious
"What did you assume that I did not tell you? Where are you least sure?"
Surfaces the silent guesses so you can check them
Try it
Ask the model what it assumed
Three free runs a day on the site key. Add your own key for unlimited runs.
That question, "what did you assume", is the most useful follow-up in the course. Models fill gaps silently and fluently. Asking them to list the gaps turns an invisible problem into a visible one you can fix in a sentence.
Which would you trust?
You asked how many days' notice you must give to end your lease. Which answer should you trust more?
Lesson FAQ
▸How do I get better answers from ChatGPT?
Ask a follow-up that names the problem and protects what worked: "too formal, keep everything else". And ask "what did you assume that I did not tell you?" to find the guesses the model made silently.
▸Why does AI give wrong answers so confidently?
It writes the most plausible continuation of your text and has no separate check for truth. Confidence is a style, not a signal. Answers that say what they do not know and point you to a source are the ones to trust.
▸Should I start a new conversation or keep going?
Keep going for one to three follow-ups; the model has useful context. If you are still not there after that, start fresh with a fuller prompt that includes what you learned.
▸Does a more expensive AI model give better answers?
Somewhat, on hard tasks. On everyday tasks the bigger gains come from a clearer prompt and one good follow-up, which are free.
Finished reading?
Mark it done to track your progress through the course.
If our calculators helped you cut down on hidden AI wallet leaks, thanks for using them. A tiny fraction of your savings is what keeps our pricing indexes updated daily.
Not sure which AI is cheapest for your use case? Find out in 30 seconds — no signup required.