AI Chatbots: The Friend You Think Out Loud With

You know that friend you call when your thoughts are all over the place? You explain the situation, bounce ideas back and forth and explore possibilities. Twenty minutes later, they haven’t necessarily given you the answer, but you understand the problem much better.

That’s a useful way to think about a chatbot. And it’s very different from how many of us first start using one.

The simplest way to use a chatbot is to ask a question and get an answer. Ask “What is the weather forecast for Brussels tomorrow?” and the AI returns a neat summary. Useful, certainly, but you’re barely scratching the surface of what the tool can do.

Now imagine asking: “I’m looking for an email app to manage my Gmail. Can you compare the best options and make a recommendation?” Something changes. Instead of simply retrieving an answer, the chatbot has to identify relevant criteria, compare alternatives and reach a recommendation.

You’re no longer simply asking AI to find something. You’re asking it to think with you.

But your new thinking buddy has an unusual habit: when it doesn’t know something about you or your situation, it may not stop and ask. It assumes.

Ask a colleague “Which email app should I use?” and they’d probably respond with questions. How many accounts do you have? Which devices do you use? What features matter most?

AI often does something different. When information is missing, it tends to fill in the blanks and produce an answer anyway.

That’s why an apparently good prompt can still produce a disappointingly generic answer. The chatbot may have answered the question perfectly; it just hasn’t necessarily answered your question.

And that’s where context becomes essential.

Imagine asking a friend for advice while leaving out half the situation. Their advice might be perfectly reasonable, for the situation they think you’re in.

Let's dive in an example with a Product Owner who needs a template for a recurring stakeholder meeting. The initial request is straightforward: “Can you create a template to help me run my bi-weekly stakeholder alignment meetings?” The chatbot creates a coherent structure, but it’s generic because the AI knows almost nothing about the actual situation.

Now give it the full story. The Product Owner works on an investment app inside a large bank. The team uses Scrum. Senior leaders attend the Sprint Review but aren’t interested in Sprint-level detail. The meeting lasts 45 minutes, with only 15-20 minutes available for slides, and the bank uses quarterly OKRs.

Same AI, same underlying problem, very different output. The resulting template can now reflect Scrum, OKRs, the available time and the audience it has been designed for.

The difference wasn’t a better model. It was better context.

need help with, the conversation will still go in circles. The same is true with AI.

We identify five core chatbot skills: problem definition, prompt structuring, context setting, output evaluation and output iteration. Prompting is only one of them.

In fact, when an AI output is poor, the problem may not be the model or even the wording of the prompt. Often, the person asking hasn’t clearly defined what they’re trying to solve.

Once the problem is clear, the prompt itself can be structured around four pillars: Role, Objective, Context and Constraints. Who should the AI act as? What are you trying to achieve? What does it need to know? What boundaries should it respect?

But even a good prompt is only the beginning of the conversation.

Probably not. You’d question it, add information, say what you don’t like or ask them to look at the problem differently. So why do we so often accept AI’s first answer?

Let's demonstrate this with a sales email. A detailed first prompt produces a perfectly acceptable B2B email: professional, inviting and easy to scan. But it isn’t particularly memorable. Instead of throwing it away, the next step is to iterate. Give the chatbot examples of emails you’ve written before and ask it to refine the draft accordingly. Examples can convey the style, structure and quality you want far more clearly than a description alone.

That’s when the relationship starts becoming genuinely collaborative. You prompt, it responds, you evaluate, you challenge, it adapts. You’re not simply extracting an answer from a machine; you’re using it as a thinking buddy.

And that’s really the shift that makes chatbots more valuable. The goal isn’t to treat the interaction as a transaction, question in, answer out, but as a conversation. Define the problem, give AI the context it’s missing, evaluate what comes back and keep iterating. The first answer doesn’t need to be the final one.

Once you start working this way, you’re also developing skills that go beyond chatbots. Problem definition, context setting, evaluation and iteration become the foundation for working with specialists, assistants and eventually agents.

So next time you open a chatbot, instead of asking “What answer can it give me?”, try asking: “How can it help me think this through?”

Eventually, however, collaboration can start to feel repetitive. Perhaps every time you ask AI to solve a certain kind of problem, you give it the same instructions, explain the same process and establish the same rules.

At some point, you have to wonder: Why am I teaching my friend the same thing every time?

What if AI could already know the process?

That’s when your thinking buddy starts becoming something else: a specialist.

And if you want to get better at this kind of collaboration before making that jump, join our Collaborating with AI Agents training to practise the foundational skills behind effective AI collaboration, from problem definition and context setting to evaluation, iteration and more advanced AI workflows.

Continue the series: AI Specialists - The Friend You Call for That One Thing