Most people use AI the wrong way. And that's a real shame.

Two colleagues, the same AI model, both wanting the same thing done. One gets back a usable piece of text and is done in five minutes. The other gets a vague, generic, meaningless answer, throws it away and concludes that AI "can't really do much anyway". Same tool. Completely different result.

I see this everywhere. And it's almost never the AI's fault. People get far less out of AI than they could, without even realising it. Such a waste. How much gets left behind has been measured in experiments, and we put those figures together separately. Because that difference isn't in the tool, it's in how you ask. And that is something you can learn.

Same model, a different result

For the past few years we've been talking about models. Which model is smarter? Which is faster? Which scores highest? That matters. But it hides something. Because for the vast majority of day-to-day work, the model has long since stopped being the bottleneck. The bottleneck is the way people steer it.

An AI model is not a search engine and not an oracle. It's a thinking partner that gets a lot better the more context you give it. Give it half a sentence and you get half an answer. Give it rubbish and you get rubbish back. That's not a shortcoming of the technology. It's a mirror. Give a capable model instructions that are too broad and it really does go wrong. We put together five examples of that.

It's a skill, not a talent

And that is exactly where the solution lies. If the result depends on how well you explain something, then it's a skill. And skills can be learned.

The thread running through all of this is simple: the better you explain, the better AI helps you. Anyone who knows how to build up a request, how to give context, how to steer when the answer isn't right yet, gets far more out of it. Not because that person is more technical, but because they have learned how to ask. That is exactly what we teach in Effective AI at Work.

This isn't about programming or clever "prompt tricks". It's a way of thinking and communicating. You wouldn't explain what you want to a bright intern in three words and then expect a perfect report. With AI, most people do exactly that. And then they're surprised when it falls short. And once such a model starts acting instead of answering, the game changes again.

Why "just giving it a go" doesn't get you far enough

"We do use AI, you know." That's often true. People open a chat window, type a question and get something back. But that's usually where it stops. They stay stuck at the level of one-off questions and miss the step towards really working with the thing. What you may and may not do with such a tool at work is a separate story, by the way.

The problem is that this plateau is invisible. You don't know what you're missing. After all, you have never seen how good it can be. The difference between "it works a bit" and "this saves me hours every week" isn't a better model. It's someone handing you the right tools. You can measure for free where your organisation stands today with the AI Adoption Scan.

And that difference isn't only in your work. The same skill that makes you more productive at the office helps you at home: drafting a letter, looking something up, wording a difficult email, helping your child with school. The same skill, at work and in your private life.

What this means for you

My point isn't that everyone has to become an AI expert. My point is that the gains are already there for the taking, with no new model and no big budget. The only investment is learning how to steer it.

For organisations this means something concrete. You probably already have the tools in-house. What's missing isn't technology, it's skill. Which tools are actually doing the rounds by now is often a surprise in itself. A few hours of focused training can lift a whole team from "it works a bit" to "this really saves us time". That is often cheaper and faster than buying yet another new tool. What that costs per person is stated openly on the pricing page. If it also has to be demonstrable under the AI Act, there is the AI literacy training as well.

That's why, at AIAdopt, we built a short, practical training around this one idea: the better you explain, the better AI helps you. No technical course, no jargon, but usable straight away. Because a lot of people use AI the wrong way. And that's a shame. There's so much to be gained. Want to see what else there is first? Look at all our training. Want to arrange it for a whole team? We can help you set it up. And otherwise, just get in touch.

Written by Rob Ummels in collaboration with Claude (Anthropic). Editorial responsibility: AIAdopt.