Cool Questions to Ask ChatGPT
Twenty prompts worth typing into ChatGPT when you want something back you can actually use: rewriting your own words, rehearsing a conversation you are dreading, turning notes into a brief, planning a way into a new skill, and finding the gaps in work you are too close to see. Each one notes what a good response looks like, and where the tool tends to fall over.
The questions
Open any question for the note
Rewrite this paragraph three ways: shorter, plainer, and funnier.
Why ask it
Working on text you wrote yourself is where these tools are most reliable, because the facts come from you. Three versions also let you pick rather than accept, and the funny one usually shows you which phrase in your original was doing no work.
Read this email before I send it and tell me how it could be misread.
Why ask it
Ask for misreadings, not improvements, or you will get a blander version of your own message. The useful output is two or three specific sentences a tired reader could take the wrong way, and the assumption you made about what they already know.
Explain this with an analogy from cooking, then one from sport, and tell me which analogy breaks down first.
Why ask it
The last clause is what makes it more than a party trick. A good answer names the point at which the comparison stops holding, which is usually the exact place where the real concept is hard, and that is the part worth reading twice.
Here is what is in my fridge and I have forty minutes. What can I cook?
Why ask it
Genuinely useful, with one caveat: quantities and timings are generated, not tested, so treat them as a starting point and use your own judgement on anything involving meat, fish or preserving. Ask it to list what you would need to buy separately.
Turn these messy notes into a one-page brief for my manager.
Why ask it
Watch the output for anything it added that you did not say, which is the failure mode here. Ask it to mark any sentence that came from inference rather than from your notes, then check those before the document leaves your desk.
Be the other person in a conversation I am dreading, and do not make it easy for me.
Why ask it
Give it their position, their likely mood and what they know. The instruction not to be easy matters, because the default will be an agreeable roleplay that leaves you unprepared. Ask afterwards which of your answers sounded weakest.
I want to learn this in eight weeks at four hours a week. What is the plan, and what should I skip?
Why ask it
The skipping is the valuable half, since anyone can produce a comprehensive curriculum. Push for what to leave out at this level, and for one thing to make each week rather than only material to read.
Ask me five questions before you answer this, then answer it.
Why ask it
The single most useful habit on this list. It prevents a confident answer to a question the model has misunderstood, and the questions themselves often show you which detail was actually load-bearing in your own problem.
Here are five books I have loved. Recommend one I probably have not heard of, and say why.
Why ask it
Expect a couple of the recommendations to be inaccurate about their contents, and check the title exists before you order it. The reasoning about what connects your five is often better than the recommendation itself.
Take an argument I am having and write the other person's case as they would sign off on it.
Why ask it
Sign-off is the standard to hold it to: nothing they would reject as a caricature. If the result is easy to knock down, ask it again and say so. Most people find their own position needs adjusting after reading a fair version of the opposite one.
Here is my CV and the job advert. What three gaps will a hiring manager notice?
Why ask it
Remove personal details before pasting anything. Useful answers point at unexplained date gaps, missing evidence for a claimed skill, and the requirement you have quietly not addressed. Ignore the generic advice about action verbs.
Interview me about my week, one question at a time, then tell me what I complained about most.
Why ask it
One question at a time is the instruction that makes this work; without it you get a list and no conversation. The summary at the end is often uncomfortably accurate, because patterns in your own week are easier for someone else to count.
Give me ten names for this, then tell me which two are usable and why the rest are not.
Why ask it
Volume is easy for a language model and judgement is not, which is exactly why you ask for both. The elimination round is where you learn your own criteria. Check any name you like for existing use before committing to it.
Write the same short scene twice: once as a picture book, once as a court transcript.
Why ask it
A quick way to see how much of writing is register rather than content. It is also a decent editing exercise on your own work: the transcript version strips out everything that was atmosphere, which shows you what the scene actually contains.
Speak to me only in the language I am learning for the next ten messages, and correct my mistakes as we go.
Why ask it
Set the level explicitly, such as beginner with present tense only, or it will pitch too high. Corrections tend to be reliable on grammar and less so on what a native speaker would really say, so treat idiom advice with more caution than conjugation.
What would an expert notice about this piece of my work that I cannot see?
Why ask it
Name the kind of expert: an editor, a structural engineer, a nurse, a tax accountant. Without that, you get generic feedback. With it, you get the two or three questions a professional would ask first, which is the part worth acting on.
What questions would a good interviewer ask me about my job, and what would the follow-ups be?
Why ask it
Interesting whether you are preparing for an interview or a podcast, and it is often a better preparation exercise than rehearsing answers. Look for whether the follow-ups actually dig, or simply restate the first question in different words.
Think of an object and let me guess it in twenty questions.
Why ask it
Worth trying because of how it fails: with nowhere to keep a secret, a chatbot may quietly change its answer to fit your guesses. It is a memorable demonstration that the thing talking to you has no hidden notepad, only the conversation so far.
Argue that I am wrong about something I have just told you, using only what I said.
Why ask it
The constraint stops it inventing counter-evidence. Because it can only use your own words, any contradiction it finds is real, and this is the fastest way to notice that two things you believe do not sit together.
What is the most interesting thing about this subject that would not be in the first paragraph of an encyclopedia entry?
Why ask it
Pushes past the summary that these tools produce by default. The answers are often genuinely good and occasionally invented, so treat any specific claim, date or name as something to check before you repeat it at dinner.
Getting useful answers out of a chatbot
Practical guidance for the conversation itself
Habits that improve almost every answer
- Bring your own material. Prompts that work on text, notes or data you provide are far more reliable than prompts that ask the model to supply the facts as well as the writing.
- Say who you are and what the output is for. "For a school newsletter" and "for a planning appeal" produce genuinely different documents; without that line you get the average of both.
- Add a constraint: a word count, a reading age, a format, a limit on what it may assume. Constraints do more for quality than any phrasing trick.
- Ask for two or three versions rather than one answer. Choosing is easier than critiquing, and the differences between versions show you what the decisions actually were.
- Tell it to ask you questions first whenever the task depends on details only you know.
When the first answer is disappointing
Say what is wrong with it, specifically
"Too long, too cheerful, and the second paragraph assumes I already have a budget" gets a better second attempt than "try again". Vague dissatisfaction produces a rearranged version of the same thing.
Ask for the reasoning, then correct the premise
If an answer is off, the mistake is usually an assumption made early. Ask what it assumed about you or the situation, correct that one item, and the rest often falls into place without further instruction.
Start again rather than arguing for long
Long threads accumulate your earlier framing and its earlier mistakes. When a conversation has gone in circles twice, open a new one and write a better opening prompt using what you have just learned.
Keep the prompts that worked
A short note of the three or four prompts that reliably help you, in your own words, is worth more than any collection of clever templates. Reuse and adjust them rather than reinventing each time.
What to be careful about
Specific facts need checking
Dates, quotations, prices, legal requirements, medical dosages, citations and page numbers can all come back wrong while sounding entirely confident. Check anything you are going to rely on or repeat, against a source you would trust without the chatbot.
Do not paste what you would not email to a stranger
Client material, medical details, salary information, other people's private messages and anything covered by an agreement at work should stay out. Strip names and identifying details when you want help with something real.
It agrees too easily
Push back on a correct answer and it may abandon it. Because that happens, its agreement is not evidence that you were right, and "are you sure?" is a weaker test than asking for the reasoning behind the original claim.
Plans and schedules are not advice
Training plans, diets, tax arrangements and study timetables come out plausible and personalised, which makes it easy to forget nobody assessed you. Treat them as drafts to take to a person who can, particularly where health or money is involved.
Praise is not feedback
Ask for what is weakest, what a hostile reader would attack, or what is missing. A request for feedback on its own tends to return an encouraging summary of what you already wrote.
A short sequence for real work
Four steps
- 1Describe the task, the audience and the constraint in one message, and end with "ask me anything you need before you start".
- 2Answer its questions honestly, including the ones that reveal you have not decided something yet.
- 3Ask for two versions, then say precisely what you prefer about each and have it produce a third.
- 4Ask what in the final version it is least confident about, then check exactly those points yourself.
The one thing to remember
These tools are strongest as an editor, a rehearsal partner and a first draft machine, and weakest as a reference book. Point them at your own material and you will get value; ask them for facts and you inherit the job of checking every one.