How to Ask AI Questions Without Making a Complete Mess of It
There is one thing I have learned since I started using artificial intelligence: AI has no idea what is going on inside our heads.
It sounds obvious, but that is where a lot of problems start.
We ask it to do something, expect it to understand exactly what we mean and, when the answer is nothing like what we had in mind, we think:
“This AI hasn’t got a clue.”
Well... sometimes it hasn’t.
But other times, the problem is that we simply haven’t told it enough.
I realised this while working on things for this very blog.
I usually knew exactly what I wanted, but I didn't always explain it properly. I would give it some instructions, it would interpret them differently, I would correct it, it would interpret something else... and eventually we could end up miles away from where I wanted to go.
And then there was London.
AI planned me a fantastic route around central London.
The small detail was that I hadn't properly explained what I considered a fantastic route.
We ended up doing more than 20,000 steps.
For a 20-year-old, probably a brilliant day.
For us, our shoes were smoking.
The route itself wasn't bad.
The question was incomplete.
And that taught me something that applies to almost anything we ask AI to do:
It isn't enough to tell AI what you want. You also have to explain what “doing it well” means to you.
And no, you don't need to learn strange technical words or become a technology expert.
You just need to learn how to use it.
1. Start by saying what you want to achieve
A lot of the time, we ask questions that are simply too open-ended.
For example:
“I want to make the best apple pie in the world.”
Great.
But what does “the best” mean?
Very juicy?Not too sweet?Lots of cinnamon?Traditional?Easy to make?For someone who isn't very experienced at baking?
AI might give you a brilliant recipe, but it might not be your perfect recipe.
We could explain it better:
“I want to make a very juicy apple pie that isn't too sweet and has plenty of cinnamon. I'm not very experienced at baking and I want a simple recipe.”
Now AI has a much clearer idea of what we're looking for.
The first question says what to do.
The second explains what we want to achieve.
And that difference is huge.
2. If you don't know what information it needs, make it ask you
This is one of my favourite tricks.
Sometimes we know exactly what we want to achieve, but we don't know what information AI needs in order to help us.
That's fine.
Let it ask.
For example:
“I want to organise a one-week trip.
Before recommending anything, ask me whatever questions you need to understand my expectations, preferences, budget and limitations.”
Instead of trying to guess which details are important, we let AI interview us.
It might ask about our budget, the pace of the trip, how much walking we are comfortable with, what we like and what we want to avoid.
And perhaps it will ask a question we hadn't even thought about.
This works for almost anything.
For buying a computer:
“I want to buy a computer. Before recommending any models, ask me everything you need to know about how I'm going to use it.”
For writing an article:
“I want to write an article about this subject. Before you start, ask me the questions you need to understand what I want to say.”
For preparing for a job interview:
“I want to prepare for a job interview. Before giving me advice, ask me whatever you need to know about my situation.”
We don't always have to know what to ask AI. We can ask AI to ask us.
3. Give it the context that could change the answer
The same question can need completely different answers depending on who is asking it.
It's not the same to say:
“What exercise can I do?”
as:
“I want to start exercising three days a week. I'm looking for something simple and I don't want a complicated routine.”
The context changes the answer.
It might be our experience, the amount of time we have, our budget, who the information is for, where we are, or any other circumstance that could make a difference.
But we don't need to tell AI our entire life story.
Give it the information that could change the answer.
4. Explain what you expect from the answer
This sounds similar to context, but it isn't quite the same.
One thing is telling AI about our situation.
Another is telling it what we expect from the answer.
For example:
“I want to organise a trip, but I don't want to try to see everything. I'd rather travel slowly and enjoy the places, even if that means missing some sights.”
That completely changes the result.
Or:
“I want a simple explanation. I don't need all the technical details; I just want to understand how it works.”
AI may know a huge amount about a subject, but it doesn't automatically know what we consider a good answer.
5. Give it limits
Here's another very simple trick.
Sometimes we assume that the more information, options or suggestions AI gives us, the better.
Not necessarily.
If we want a quick recipe:
“I don't want it to take more than 30 minutes.”
If we want recommendations:
“Give me only five options.”
If we're preparing a piece of writing:
“I want around 800 words.”
If we're planning a trip:
“I don't want to walk more than 8 kilometres a day.”
Limits make the answer much more realistic.
AI can't know where our limits are if we don't tell it.
6. Tell it what you DON'T want too
We spend a lot of time explaining what we want and forget to say what we want to avoid.
That is information too.
For example:
“I want an apple pie recipe, but I don't want it to be too sweet.”
Or:
“I want a quiet hotel, but I don't want to be isolated from the centre.”
Or:
“I want you to improve this text, but I don't want you to change the way I write.”
We can also say:
“Don't make up information.
If you're not sure about something, tell me.”
Telling AI what you don't want can be just as important as telling it what you do want.
7. Give it an example
Some things are difficult to explain with words.
And sometimes one example is worth ten explanations.
We can say:
“I want something like this example, but adapted to my situation.”
If we want a particular writing style, showing AI a piece of text can be much easier than trying to describe exactly how we want it to write.
If we want a particular type of recipe, we can show it one we like.
If we want a simple explanation, we can give it an example of the kind of explanation we understand.
It isn't about copying it.
We're showing AI what we mean when we say “I want something like this.”
8. Before it starts, check that it has understood you
This trick is particularly useful when the task is long or important.
We can say:
“Before you start, tell me how you have understood my request.”
AI will explain what it thinks we are asking it to do.
And that gives us a great opportunity.
If it has understood correctly, carry on.
If it has understood something else, correct it before it starts working.
It is much easier to say:
“No, that's not what I meant. I want to keep X and change Y.”
than to receive three pages of something we never asked for.
9. You can have a conversation with it, but don't lose the plot
One of the advantages of AI is that we don't have to get the perfect question right first time.
We can start with an idea and refine it.
For example:
“I want to plan a trip to Australia.”
Then:
“But I don't want to see everything.”
Then:
“I'd rather travel slowly.”
Then:
“And I want to avoid driving for too many hours.”
That can work brilliantly.
But be careful.
AI can be too helpful.
You start talking about flights and suddenly it's recommending hotels, excursions, restaurants, suitcases, insurance, beaches and, if you let it carry on, probably dessert for dinner as well.
The conversation can start wandering off.
If that happens, we can say:
“Let's go back to the original question. Focus only on the goal I gave you.”
Or:
“Don't add recommendations that aren't related to my question.”
And when a conversation has turned into a mixture of five different subjects, sometimes the best thing to do is start a new chat.
A longer conversation isn't necessarily a better conversation.
10. Don't hand over the work: use AI to do it better
I think this is one of the important problems of this new era.
AI can write your CV, prepare a presentation, do your homework, draft a letter, organise a project or even prepare the answers for a job interview.
And that is precisely where the danger lies.
Just because it can do something doesn't mean we should let it do everything.
If we ask AI to do the work and hand it in without reading it, we've saved time, yes.
But we've also stopped thinking, learning and checking.
And then something rather ridiculous happens, we use a tool to do our work, don't check the result and, when something goes wrong, we blame AI.
AI shouldn't replace our judgement.
It should help us use it better.
We can use it to:
give us ideas when we're stuck;
organise what's in our heads;
review a text we've written;
suggest another structure;
find weak points;
explain something we don't understand;
give us alternatives;
spot possible mistakes;
ask us questions we hadn't thought of;
take care of repetitive work.
But afterwards we read, decide, correct and make the result our own.
A very simple example
Let's say we want to prepare for a job interview.
We can ask:
“Give me the questions I might be asked in an interview for this position.”
Perfect.
Then:
“Interview me and tell me where I'm weak.”
Also perfect.
But if we say:
“Prepare all my answers and I'll memorise them.”
Now we have a problem.
Because on the day of the interview AI won't be sitting next to us.
And if they ask us something different from the answers we've memorised, we may discover that the problem wasn't that AI didn't know how to answer.
The problem was that we didn't know how to answer.
This applies to studying, working, writing, preparing presentations or practically anything else.
AI can do the work for you.
But if you always let it do it, eventually you may not know how to do it yourself.
We don't need to demonise technology.
But neither should we turn it into a crutch for everything.
AI is a tool to help you, not a replacement for your brain.
11. If it gets something wrong, correct the specific mistake
You don't have to start again from scratch every time AI makes a mistake.
Instead of:
“That's wrong.”
We can say:
“Point 3 isn't correct. I want X, not Y.
Keep everything else the same.”
Now it knows exactly what needs changing.
And this is also part of learning how to use AI.
We don't need to get angry because it made a mistake, but we shouldn't accept an incorrect answer just because it sounds very confident.
12. If something is important, check it
Now we come to a rule I consider essential.
AI can be extremely useful and still get things wrong.
It can give us an incorrect date, an outdated price, opening hours that have changed, the wrong address, or a piece of information that sounds completely real but isn't.
So if the information is important, check it outside AI.
There are several ways to do this.
Ask another AI
Ask the same question in another system and compare the answers.
If they are different, that's a signal to investigate.
But be careful:
Two AIs giving you the same answer doesn't mean it's true.
They can both be wrong.
Search the web
And, whenever possible, go directly to the source.
Check a specialist website
Depending on the subject, it may be much more reliable to consult:
an official website;
a government body;
the manufacturer;
an institution;
a specialist publication;
the official website of a museum, transport company or destination;
a professional source related to the subject.
The more important the information, the more important it is to check it against the right source.
For choosing an apple pie recipe, we probably don't need to launch an international investigation.
For a regulation, an amount of money, a booking, opening hours, a legal question or anything that could have consequences, I wouldn't rely on an AI answer alone.
13. The final trick: ask what you've missed
This may be one of the most useful tricks of all.
Because we can make a huge effort to give AI context, expectations, limits and examples...
and still forget something.
That's human nature.
So once we've explained what we need, we can ask:
“Is there anything important I haven't told you that could change your answer?”
Or:
“What information are you missing that would help you give me a better answer?”
Or, after receiving the answer:
“Which parts of this answer should I check or verify before I rely on it?”
We're using AI not just to answer us, but also to help us discover what we hadn't thought about ourselves.
And that works for almost anything.
Maybe we're making an apple pie and forgot to mention that we have a small oven.
Maybe we're buying a computer and forgot to say that we'll be carrying it around every day.
Maybe we're planning a trip and forgot to say how much walking we're prepared to do.
We don't always know what information we're missing.
But we can ask.
🤖 AI Answers
Do I need to learn complicated prompts?
No.
In fact, if someone starts talking about “prompts” and makes you think you need to learn a new language to use AI, don't overcomplicate it.
What's important is knowing how to explain what you want to achieve, providing the context that could change the answer and saying what you expect.
And if you don't know what information it needs, simply say:
“Ask me whatever you need to know before answering.”
If I ask the wrong question, does AI get angry or penalise me?
No.
If the question is too open or doesn't contain enough information, it may give you an answer that doesn't fit what you were looking for.
Simply rephrase it, add some information or tell it what it misunderstood.
There's no need to be afraid of correcting it.
How do I know whether what it tells me is true?
You can't know simply because the answer sounds convincing.
If it's an important fact, check it.
You can ask another AI, search online and, above all, check a reliable or specialist source.
So what's the point of asking several AIs?
To compare.
One may explain something in a way you understand better. Another may give you an option you hadn't considered.
And if their answers are very different, that may be a sign that you need to investigate further.
But comparing doesn't replace checking.
Do I have to tell it absolutely everything?
No.
It's not about writing a novel every time you want to ask a question.
It's about giving it the information that could change the answer.
If you're asking about a recipe, it may matter whether you want it to be quick, what ingredients you have or whether there's something you want to avoid.
But you don't need to tell it your entire life story from the day you were born.
And what's the most important trick?
Probably this:
If you don't know what information AI needs, make it ask you. And when it has finished answering, ask it what you might have missed.
Because the best question isn't always the one we know how to ask.
Sometimes the best question is asking AI to help us find the question we don't yet know how to ask.
AI isn't perfect either
Sometimes we talk about artificial intelligence as if it were some kind of electronic brain that should know everything.
It doesn't work like that.
Behind these systems there are enormous amounts of information and a huge amount of human work.
People create data, evaluate answers, identify mistakes and help improve the models.
And we, as users, are also constantly learning what works, what doesn't and how we can use these tools better.
That said, an AI doesn't necessarily learn from every conversation in real time or automatically turn every user's mistake into knowledge for everyone.
It depends on the system and on how it has been designed and trained.
But there is something we can do, learn from our own mistakes when using it.
If we ask it something and the answer is wrong, we can ask ourselves:
Was it missing information?
Did I fail to explain what I expected?
Did I leave out an important limitation?
Did I give it too many things at once?
Or did it simply make a mistake?
That last possibility always exists.
And perhaps that's the most sensible way to use artificial intelligence.
Don't expect it to be infallible. Learn how to work with it.
We bring the experience, judgement, preferences and decisions.
AI brings speed, ideas, the ability to organise information and another way of looking at a problem.
And the better we learn to work with it, the more useful it becomes.
Your quick cheat sheet
Before you hit “send”, ask yourself:
What do I want to achieve?
What does AI need to know about my situation?
What do I expect from the answer?
What limits do I have?
What do I want to avoid?
Can I give it an example?
Do I need it to ask me questions before starting?
Has it really understood what I'm asking?
Are we keeping the conversation focused?
Do I need to check any information before relying on it?
And above all, remember:
You don't need to learn to talk like a machine. You need to learn how to explain to the machine what you need.
And if you don't know what it needs to know...
make it ask you.
Conclusion
In the end, learning to use AI isn't about learning strange words or knowing how to write the perfect prompt.
It's about learning how to explain to a tool what you need and, above all, not stopping thinking while you use it.
You can ask it for ideas, get it to organise what's in your head, review a piece of writing, suggest alternatives or even ask you questions before it starts.
But the work is still yours, you decide what you want to achieve, you check that the answer makes sense and you decide what is useful and what isn't.
Because there is a dangerous temptation, give AI the whole job and accept the result without reading it.
And then, when something goes wrong, it's very easy to blame the machine.
But if we haven't even checked what it has prepared, perhaps the problem isn't that AI worked badly.
Perhaps we stopped working.
AI is a tool to help us, not a replacement for our judgement. And the better we learn to use it, the more useful it becomes.
In short:
Explain what you need and give it the relevant context.
If it doesn't know something, make it ask you rather than letting it guess.
Give examples when it's difficult to explain what you're looking for.
Break large tasks into smaller parts and review each one.
Don't be afraid to correct it and ask again.
If the information matters, check it with another AI, on the web or with a specialist source.
And, above all, don't let AI think for you. Make it work with you.
Don't let AI think for you. Make it work with you.
If you think this might help someone, feel free to share it.
Comments