I've sat with a few thousand people while they learn AI. The ones who get good fast all do the same unglamorous thing, and it's almost never the thing they expected.
Here is what I tell people in workshops. Most of you will use AI to learn by asking it to explain a topic, then nodding along and forgetting it by lunch. I’ve done it myself. The version that actually works is duller and harder: name the skill and your level, ask for one concept at a time, then make the AI quiz you and tell you where you’re wrong. Add a short Friday review of the week. That loop is the whole trick, and it’s the part people skip.
I’ll say the awkward part first. AI is not as good as a brilliant human teacher. It can’t read your face, it doesn’t lie awake worrying about whether you pass, and every so often it will tell you something flatly wrong in the calmest possible voice. So why lean on it to learn anything?
Because the magic of a good tutor was never the fact that they were human. It was that they adapted to me. A good tutor explains something, watches me not get it, and tries a completely different angle. They’ll answer the dumb question I’d never raise my hand to ask in a room of twenty people. That adaptiveness, that willingness to go round again, is the one thing a language model genuinely does well.
There’s a bit of early research that points the same way. In a semester-long study at UniDistance Suisse, psychology students who actively worked with a personal AI tutor came out roughly 15 percentile points ahead of a parallel group that didn’t.[1] Small sample, one study, so I wouldn’t bet the house on the exact number. But it matches what I see week after week with the people I train.
The word doing all the work in that study is “actively.” When I watch someone treat AI as a thing to read, they get a chattier textbook and not much else. The learning only shows up when you force it to react to you. That’s the whole game, and the rest of this is just how to set it up.
You don’t need anything clever. ChatGPT, Claude, and Gemini all do this job, and all three have a free tier that’s plenty to start. Honestly, open whichever one you already have a tab for. If these tools are still new to you, our plain-English guide to AI literacy is a five-minute warm-up before you dive in.
The setup is where people win or lose, not the tool. Left to its defaults, AI behaves like that colleague who answers the question you asked and then keeps going for three more paragraphs. For learning you want a tutor who slows down, checks you’re following, and circles back. So you have to tell it to be one. It won’t volunteer.
Open a fresh chat and paste something close to this:
“Be my tutor for [skill]. My level right now is [beginner / I know the basics / intermediate]. My goal is [be specific: ‘build a basic budget in Excel’ beats ‘learn Excel’]. Teach me one concept at a time. After each one, ask me a short question to check I got it, and wait for my answer before you carry on. Keep explanations under 150 words unless I ask for more.”
That one instruction changes the whole experience. Instead of a wall of text, you get a lesson the size of a human attention span, then a check. If you’re in ChatGPT, drop it into your custom instructions so you never type it again. Our custom instructions setup guide shows the two minutes of clicking that saves.
Here’s the bit that stings, because I spent years getting it wrong myself. Reading something twice feels productive and does almost nothing. What actually moves a fact into your long-term memory is dragging it back out of your own head when it’s gone a bit fuzzy. Learning researchers call it active recall, and it’s about as settled as anything in the field gets.
AI happens to be a very good machine for forcing it on you. The loop I run with people looks like this:
The gap between this and re-reading is bigger than it looks. When you grind for an answer, miss, and then see the correction, your brain quietly flags that thing as worth keeping. Read the right answer cold and it slides straight off. The little jolt of being put on the spot is the point, not a side effect.
If the questions feel comfy, they’re too soft. Tell it: “Make these harder, and throw in ones that force me to connect two ideas.” You don’t really know something until you can use it next to something else.
Most of the value lives in five or six prompts you reuse forever. Keep them somewhere you can paste from. If you want to get sharper at writing prompts in general, our 4-part prompt formula is the obvious next read.
To test yourself: “Quiz me on what we’ve covered. One question at a time, wait for my answer, tell me if I’m right before the next.”
To find your gaps: “Based on my answers so far, what have I clearly got and what am I still shaky on? Give me one more question on my weakest bit.”
To get unstuck: “I’m not following [concept]. Try it three ways: a plain analogy, a concrete example, and step by step.”
To make it real: “Give me a small, realistic task that uses this, then review my attempt and tell me what to improve.” This is the prompt that turns knowing-about into can-actually-do, and it’s the one most people never reach for.
To space it out: “Pretend a week has passed. Quiz me on last week’s material and see what stuck.” Coming back to something after a delay is the other half of making it permanent.
Look at those again and notice none of them ask for an explanation. Every one makes you do something. Internalise that and you can throw the rest of this away.
There’s one more I keep in my back pocket, and it feels slightly daft the first time you try it: “I’m going to explain [concept] back to you. Listen, then tell me where I’m wrong or thin.” Teaching a thing is the hardest test of whether you actually have it, and the AI is a patient audience that won’t smirk when you stall halfway through your own sentence. The exact spot where you trail off is the gap you go back and fix. I use this at the end of a topic, right before moving on, as a last gut-check that it landed instead of just feeling like it landed.
The best study plan is the one you’ll still be doing in three weeks. So bin the fantasy of two-hour deep-work sessions every evening. Nobody I know keeps that up past a Wednesday. Here’s the rhythm I actually recommend.
Monday to Thursday, twenty minutes. One short session a day. One or two small concepts, run through the loop above, and finish by having the AI quiz you on what you just did. Twenty honest minutes beats two distracted hours, mostly because you’ll come back tomorrow.
Friday, fifteen minutes. Nothing new. Get tested on the whole week. You’ll be a bit embarrassed by what’s already slipped, and clawing it back is exactly what locks it in for good.
Once a week, do the actual thing. Not a hypothetical. Learning Excel? Build a sheet you genuinely need this week. Learning to write better? Draft a real email and let the AI tear into it. A skill lives in the doing, never in the explaining.
If you’d like that shape laid out as a full timeline for AI itself, our 30-day plan shows what a sane ramp looks like. Small daily reps plus a weekly review will carry you through almost any skill, not just this one.
One habit makes the whole thing stickier, and it’s barely any effort. Keep a scrappy log of what you’ve learned. Not a neat document, just a note where you jot the one or two things that finally clicked and the questions you got wrong. You can even have the AI write it for you: “In three bullets, what did I learn today and what should I review next time?” The log doubles as your Friday revision sheet, and on the days learning feels like wading through treacle, which is most days, it’s the only proof you’re moving at all. That proof is usually what keeps people going long enough to get good.
A handful of failure modes come up over and over in the rooms I teach. Worth naming them so you can see yourself doing them.
Mistaking reading for learning. The big one. You ask for an explanation, get a gorgeous summary, feel clever, and three days later it’s gone. No test, no learning. Build the quiz into every single session or you’re just browsing.
Letting it sprint. AI wants to be thorough, so left alone it’ll cram ten concepts into one reply. Keep telling it to slow down to one at a time. When you’re starting out, going deep on one idea beats skimming five, every time.
Living in the chat window. You can read about swimming for a month and still sink. At some point you have to attempt the thing, badly, and get notes. The chat is scaffolding around the practice, not a substitute for it.
Letting it think for you. This is the quiet trap. Hand the AI your code, your essay, your analysis, and you’ll end up with the output and none of the skill. Use it to coach your attempt. Don’t let it replace the attempt.
One honest warning before you go all in. These tools sometimes state things that are simply wrong, in the same confident tone they use for the things that are right. When you’re learning, that matters more than usual, because a tutor that teaches you something false has just installed a bug you’ll happily repeat.
For most everyday skills, this barely registers. The basics of Excel, project planning, clear writing, that’s well-trodden ground and the AI handles it reliably. The risk climbs the more technical, niche, or recent the topic gets. If being wrong actually costs you something, cross-check the load-bearing facts against a source you trust.
A habit I’ve picked up: when the AI says something surprising or something a lot rests on, I ask “how sure are you about that, and how would I check it?” It won’t always know, but it’ll often own up to where it’s standing on thinner ice. Our guide on how to fact-check ChatGPT covers the quick checks worth building into your routine.
None of this is a reason to back off. As a coach that tests you, adapts to you, and never once sighs at your questions, AI is about the best learning tool most of us have ever had within reach. You just have to keep your hands on the wheel. And if you’d rather learn AI itself with some structure and a person to ask, that’s exactly what our AI courses for non-technical professionals are built for.
Yes, as long as you keep it interactive. The trap is asking for explanations and reading them passively, which feels like progress and isn’t. Set it up as a tutor that teaches one concept at a time and quizzes you on each, because the testing is what makes it stick. Run that way, it’s close to having a patient tutor on call.
ChatGPT, Claude, and Gemini all do the job and all have a free tier that’s enough to start. The tool barely matters here. How you set it up matters a lot. Pick the one you already use, then tell it to go slowly, teach one concept at a time, and test you after each.
Depends on the skill, but your routine decides more than your hours do. Twenty focused minutes a day, four days a week, with a short Friday review, beats the occasional marathon session by a distance. On a practical skill, most people feel real movement within a few weeks of keeping that up.
Reading instead of recalling. AI hands you clear, confident explanations that feel like learning, but anything you only read tends to evaporate. The fix is active recall: get quizzed, answer from memory, get corrected. That small jolt of being put on the spot is what moves things into your head and keeps them there.
No, and you shouldn’t. It can state wrong things confidently, which is risky when you’re learning because you’ll repeat the error. For common skills the odds of trouble are low. For technical, niche, or recent topics, check the key facts against a trusted source, and just ask the AI how you’d verify anything that feels important.
This guide was written by Sana Mian, Co-Founder of Future Factors AI, drawing on her work as a learning designer training 2,000+ non-technical professionals to learn and apply AI. Study findings cited reflect published research available as of June 2026.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.