AI has been around for over a decade, but November 2022 was a step change. ChatGPT put a conversational AI assistant in everyone’s hands, and almost overnight it went from a niche tool to something on every desk.
I have been using it personally and professionally since then. Personally, it has been genuinely brilliant: holidays planned, children’s stories written, DIY projects sorted, code that actually works. At work, something different happens. People try it, get a mediocre answer, and decide the hype is overblown.
I do not think that is a technology problem. It is a behavioural one. The model is not the issue. The way we approach it is. These are the five habits that make the difference.
What are the five tips for working with AI?
Read them in order. They build on each other, from how you approach the tool, through how you talk to it, to who stays responsible for what comes out.
Be curious and play.
Fluency comes from use, not from reading about it.
Humans are mammals. Mammals learn by play. It sounds simple, and it is. But most people do not bring that playful energy anywhere near their work AI.
Go down the rabbit holes. Ask it something odd. Give it a task where it does not matter if it fails. Some of those rabbit holes will go nowhere, and that is fine. That is exactly where you learn the shape of what you are dealing with. The people who are genuinely good at this are not the ones who waited for a training session. They are the ones who spent time playing. Bring the same curiosity you use at home into the office. The tool is the same.
Context is king.
The model knows language. It does not know your world. Tell it.
“Context is king” is not a new phrase. But it has never been more relevant than it is right now.
Think about what happens when you walk up to a colleague and say: “Can you review this?” They are going to have questions immediately. Review what, exactly? How do you want me to review it? What kind of feedback are you after? When do you need it? You would not think twice about giving that context to a person, but we routinely fire questions at AI with none of it. Ask without context, and you will get an answer without context. It is that direct.
The model knows language. It does not know your audience, your deadline, your house style, or what good looks like to you. So tell it. This is also where my day job leaks in: the words you feed a model set the ceiling on what it can give back. It is the same instinct behind the Data Guiding Principles for a data team, scaled down to a single prompt.
It's a conversation, not Google.
The value is in the second and third turns, not the first answer.
Google trained us to fire a query and take the first result. One shot, move on. That habit is expensive when you bring it to an AI assistant.
The value is in the interaction and the iteration, the to and fro. That back-and-forth is not inefficiency, it is how the context grows, and context is what drives the quality of the answer. Correct it. Tell it what you actually meant. Ask it to try again. Ask what it missed. The second and third turns are where the quality lives, and they cost you a sentence each.
Don't live in a bubble.
Don't marry one tool, and don't let it only ever agree with you.
Most AI products are built to be agreeable. That is not an accident. Agreeable feels useful, and validation is comfortable. But there is a real difference between an acquaintance who tells you what you want to hear and a trusted friend who gives you the honest version when you ask for it.
If you only ever ask AI to write things better, you will get better content. If you ask it to challenge your thinking, you can get better decisions. The distinction matters. Ask it whether you have missed something. Ask it to argue the opposite case. Tell it to be adversarial. These are not niche prompting tricks. They are what turns a tool that flatters into one that actually sharpens your thinking.
There is a second bubble too: the tool bubble. The leading models leapfrog one another every few months. Keep a second assistant to hand and run important questions past both.
And it is not only about which model is ahead this month. Each one thinks in its own way, so on the decisions that really matter it is worth running them past two models independently. You get different considerations, different reasoning, and often the two will disagree. Those tension points are the gold. They are exactly where you should focus, because that is where the thinking that makes a difference tends to live.
Keep a human in the loop.
The AI is not accountable for the output. You are.
AI can hallucinate. This is not a secret and it is not a flaw that is about to disappear. It will state wrong things with total confidence and no warning: a figure that is slightly off, a quote that was never said, a fact that sounds right and is not.
Think of it like a close friend who can embellish. You enjoy the conversation, you benefit from the exchange, but you do not blindly take everything they say as fact. Sometimes you check for yourself. With AI, that “sometimes” should be “often.”
Stay in the loop. Use it to get to a strong draft fast, then spend the time you saved on the part only a human can do: the check, the judgement, the call. Read every line it gives you as if you wrote it, because once it leaves your hands, you did.
Putting it together
One of the most interesting things I have found on this AI journey is not about what the technology can do. It is about what gets the most out of it. And the answer, consistently, is interacting with it more like you would a human.
That does not mean treating it as sentient. It means applying the same instincts that make any good working relationship work: curiosity, context, conversation, honesty, accountability. Be curious and play. Lead with context. Have the conversation rather than firing the query. Push back rather than accepting easy agreement. And keep yourself in the loop, because the accountability never leaves you.
The same logic scales beyond a single conversation. What is true for one person and one prompt is true for an organisation and its data: a model is only ever as good as the context and the definitions you feed it. That is the whole argument behind the Data Guiding Principles and the Decision Outcome Framework.
Frequently Asked Questions
How do I get better results from AI assistants?
Treat it as a skill you build by using it. The biggest gains come from four habits: give the model the context it cannot know on its own, work in a back-and-forth conversation rather than firing a single query, try more than one tool because they leapfrog each other constantly, and keep a human reviewing and owning the final output. The quality of the answer is usually set by the quality of the input, not by the model.
What does "context is king" mean when prompting AI?
A large language model knows language, but it does not know your audience, your constraints, or what a good answer looks like in your world. Context is everything you add to close that gap: who the output is for, what it is being used to decide, the tone you need, and an example of good when you have one. The single fastest way to improve an AI answer is to add the context you were assuming it already had.
Should I trust what an AI assistant tells me?
Trust it the way you would trust a fast, confident, occasionally wrong junior colleague. It is brilliant for a first draft, a second opinion, or a starting structure. It is not a source of truth. It will state wrong things with total confidence. Check facts, figures, and quotes before they leave your hands, because the moment you ship it, it is yours, not the model's.
Which AI assistant is best, ChatGPT, Claude, Copilot or Grok?
There is no permanent winner, and that is the point. The leading models trade places every few months, and each has tasks it is better at. Pick a primary that fits your workflow, but keep a second one to hand and run the same question past both when it matters. Loyalty to one tool is the fastest way to fall behind.
What does "human in the loop" mean?
It means a person stays responsible for reviewing, deciding, and owning anything the AI produces. The model accelerates your judgement, it does not replace it. Used well, AI does the first eighty per cent so you can spend your attention on the twenty per cent that needs a human: the check, the nuance, and the accountability.
The same instinct, scaled to a data team.
Context and clear definitions decide what a model can give back. The Data Guiding Principles are how that belief becomes the way a data and analytics team actually works.
Read the Data Guiding PrinciplesMark Cunningham is the founder of Insights². He has spent more than a decade building data and analytics products, and rather too many recent evenings finding out what these tools can and cannot do. He writes at insights-2.com about how to make the meaningful measurable.