Learning Faster With AI: A Realistic Look
What generic AI tutoring actually gets right and wrong when you're trying to learn something hard, four prompts that speed up real learning today, and what changes when the AI applies Barbara Oakley's documented learning-science framework instead of generic study advice.
Ask AI to teach you something and it will, clearly and patiently, for as long as you want. That's also close to the least effective way to use it for learning, because a good explanation produces a strong feeling of understanding that doesn't reliably predict what you can actually recall later, a well-documented gap between feeling like you've learned something and having actually learned it.
The more useful use of AI for learning isn't explanation. It's generating the practice and retrieval that explanation alone can't provide, which is a less comfortable but far more durable way to actually get better at something.
What actually happens when you ask AI to teach you something?
Ask an AI to explain a hard concept and it will produce something genuinely clear, often clearer than the textbook you were struggling with. It feels like real progress. The problem is that the feeling of understanding a clear explanation gives you is a documented poor predictor of what you'll actually be able to recall or apply later, once the explanation isn't in front of you anymore.
This isn't a flaw in the AI's explanation. It's a mismatch between the activity (reading, recognizing, nodding along) and the activity that actually builds durable memory (retrieving the information yourself, under some difficulty, without the answer in view), and no amount of explanatory clarity closes that particular gap.
Where does AI actually help you learn, and where should you doubt it?
Unaided, it's genuinely strong at generating fresh practice problems on demand, endlessly, at exactly the difficulty level you specify, and at explaining the same concept multiple different ways until one framing clicks for you specifically.
It's weak at telling you whether you've actually learned something versus merely recognized it while it was explained to you. A confident-sounding answer to a question about material you half-remember can look identical to real understanding until you're tested without the text in front of you, at which point the gap between recognition and recall becomes obvious, usually at an inconvenient moment.
What four prompts actually speed up real learning?
These four work in Claude, ChatGPT, or Gemini as written.
1. The retrieval quiz. "Quiz me on [topic] with five questions, one at a time, don't show me the next question until I've answered the current one, and tell me if I'm wrong before moving on." Why it works: this forces retrieval, producing the answer from memory rather than recognizing it in a text, which is documented to build far stronger retention than re-reading the material again.
2. The interleaved practice set. "Give me ten practice problems on [topic], but mix in problems from [a related topic I've already covered] randomly, don't group them by type." Why it works: mixing problem types (interleaving) forces you to first identify which method applies before applying it, which is closer to how you'll actually need to use the skill later than practicing one method in a block, where the method is already given away by the fact that every problem in the block uses it.
3. The explain-it-back check. "I'm going to explain [concept] back to you in my own words. Tell me specifically what I got wrong or oversimplified, don't just tell me if I'm generally right." Why it works: producing an explanation yourself is a stronger test of real understanding than recognizing a correct one, and asking for specific errors rather than a general verdict catches the parts you've quietly glossed over without noticing you'd done it.
4. The spaced review generator. "Generate three new questions on [topic I learned a week ago], different from any I've seen before, to check what's actually stuck." Why it works: spacing practice out over time, rather than massing it into one session, is one of the more robustly replicated findings in learning research, and fresh questions each time prevent you from just memorizing the specific question rather than the underlying concept.
Where does this still fall short?
Run all four regularly and you'll retain material noticeably better than passive reading or one-off AI explanations alone. What you won't have is a structured curriculum, a sequence that tells you what to learn in what order, when you're actually ready to move on, and when a gap in an earlier topic is quietly sabotaging a later one that depends on it. The gap isn't the practice technique; it's the overall sequencing a real course or curriculum provides deliberately and an ad hoc set of prompts doesn't.
How does Barbara Oakley's framework actually change how you use AI to study?
Barbara Oakley, drawing on her own documented work in A Mind for Numbers and subsequent research, built chunking into a specific, learnable skill: compressing a pattern or procedure into a single, fluently retrievable unit you can recall and apply as one move, rather than reconstructing it from first principles every time. She pairs it with documented findings on interleaving and spaced practice, and treats focused, effortful practice and diffuse, unfocused rest as two modes that both matter, not just one.
Before, generic prompting: asked for help learning a new programming concept, a generic AI response explains it clearly with an example, then offers to explain further if anything's unclear, treating explanation as the whole job.
After, Oakley's framework applied: the framework treats the explanation as only step one. It follows immediately with a small set of graduated retrieval problems specifically designed to force chunking, starting with a problem that requires recognizing the pattern in isolation, then a harder one that requires recognizing it mixed in with a previously learned, easily confused pattern (interleaving), and finally schedules a follow-up quiz for several days later rather than treating the session as complete once you've gotten a problem right once. The framework's documented position is that getting a practice problem right immediately after an explanation is a weak signal; getting it right after a delay, without the explanation freshly in mind, is the actual test.
A generic study-help prompt has no built-in reason to run the full sequence, explain, retrieve, interleave, space it out, unless something specifically requires it to. The framework does, every time, regardless of how clear the initial explanation already felt.
How do you actually get this framework running in Claude, ChatGPT, or Gemini?
Claude installs it as a Skill: Settings → Skills → Add skill, upload the .zip. It auto-invokes when your question matches the skill's description, or gets forced directly with /barbara-oakley-framework. In Claude Code, the same package sits in ~/.claude/skills for every project, or a project's own .claude/skills folder for just that one.
ChatGPT has no Skills equivalent, so the plain .md becomes a Custom GPT's Instructions: Create a GPT → Configure → Instructions, pasted in directly. Skip the default Custom Instructions fields, they cap at 1,500 characters; a Custom GPT's Instructions field runs to roughly 8,000, enough for the full framework.
Gemini has no dedicated upload option, so the content goes into the system prompt if your interface has one, or opens the first message instead: "Here is a thinking framework to apply throughout this conversation. Read it carefully, then answer using this framework's approach," with the framework text pasted underneath.
What should you actually do with this next?
Oakley's documented method is one of several in the Educator category, alongside Sal Khan and John Dewey, each an .md file readable by Claude, ChatGPT, or Gemini, worth exploring directly if learning how you learn is the actual problem you're solving for, rather than one specific study session.
Frequently asked questions
Is it cheating to use AI to help me learn something?
Not for the use covered here. Asking AI to explain a concept differently, generate practice problems, or quiz you isn't outsourcing the learning, it's structuring your own effort more effectively, the same way a good tutor or textbook does. The line worth watching is different: having AI produce the finished output you're supposed to be learning to produce yourself, an essay, a proof, a working program, defeats the purpose in a way that asking it to explain or quiz you doesn't.
Why do I forget things I 'learned' from an AI explanation almost immediately?
Because reading a clear explanation feels like learning but isn't the same activity as retrieving the information yourself, which is what actually builds durable memory. This is documented, not a personal failing: passive review, including a great AI explanation, produces a strong feeling of understanding that doesn't reliably predict what you can actually recall later without the text in front of you.
What's the biggest mistake people make using AI to study?
Asking it to explain instead of asking it to quiz. Explanation-seeking feels productive and produces weak retention; being tested, and getting it wrong, then correcting the error, produces much stronger retention, a well-documented finding sometimes called the testing effect. Most people default to the less effective mode because it's more comfortable and produces less of the frustration that real retrieval practice tends to involve.
Can AI replace a real course or a real teacher?
Not for structure, sequencing, or accountability, all of which a well-designed course provides deliberately and an ad hoc AI conversation doesn't provide on its own. It's genuinely strong as a supplement, for on-demand explanation, generated practice, and quizzing, alongside a structured course or curriculum, rather than as a full replacement for one, especially for a subject with real prerequisite structure that's easy to skip past without noticing.
How is chunking different from just memorizing more?
Chunking is building compressed, retrievable units, a fluent grasp of a pattern you can recall and use as a single move rather than reconstructing from parts every time, the way a chess player recognizes a board position rather than calculating every piece from scratch. Barbara Oakley's documented work treats it as a distinct skill from raw memorization, and one AI-generated practice can specifically be built to train.
Does spaced repetition actually work, or is it oversold?
It's one of the more robustly documented findings in learning science specifically, tested across many subjects and skill types, not oversold folklore. The core finding, that spacing practice out over time beats massing it into one long session, holds up consistently; what AI adds is the ability to generate fresh practice material for each spaced session instead of running out of new questions after the first pass.
What if the subject I'm learning doesn't have obvious 'chunks' to build, like a soft skill?
Most skills have more structure than they first appear to. A soft skill like negotiation or public speaking still has recognizable component moves, a calibrated question, a specific opening technique, that can be practiced and chunked individually before being combined into the fluid whole, even though the chunking is less obvious than in math or a language, and naming those component moves is itself half the work.
What's the honest limit of using AI to learn faster?
It has no way to verify you're not fooling yourself about how well you actually know something, since a confident answer to a question you half-remember can pass for real understanding until you're tested cold on it later. Use it to generate the retrieval practice that actually catches that gap; don't rely on the comfortable feeling of a good explanation as proof you've genuinely learned it.
Written by Gareth Hoyle. Last updated 26 September 2026. Part of the authority.md guides library.
More guides.
Bringing AI Into a Creative Project Without Ruining It
What generic AI brainstorming actually gives a creative project and what it quietly flattens, four prompts that help today without taking over, and what changes when the AI applies Rick Rubin's documented listening practice instead of generic idea generation.
AI for Hiring Decisions: Where Claude and ChatGPT Actually Help
A grounded look at what generic AI use gets right and wrong when you're deciding on a hire, four prompts that genuinely sharpen the decision, and what changes when the AI applies a documented evidence-based hiring framework instead of generic interview advice.
Using AI to Prep for a Negotiation: What Actually Works
Generic AI negotiation advice is genuinely useful and genuinely limited. Here's what Claude and ChatGPT get right before a negotiation, four prompts that work today, and what changes once the AI has a documented framework instead of a generic one.