--- title: "Claude Learning Mode: What It Is and How to Use It" canonical: https://aaliyaan.com/blog/claude-learning-mode-guide/ pubDate: 2026-06-19T00:00:00.000Z author: "Aaliyaan Chaudhary" description: "Claude learning mode has two versions, one for studying and one for coding. Here's what each does and when to actually turn it on." tags: ["claude","ai","developer-tools","education","ai-dev-tools"] --- Claude learning mode is not one feature. It is two separate ones that share a name and a Socratic teaching style. One lives in claude.ai as a style you pick for any conversation. It guides students through a problem with questions instead of handing over answers. The other lives in Claude Code as an output style for programmers, and it makes Claude leave parts of the code for you to write yourself. I have used both, and confusing them is easy to do since Anthropic named them almost identically. Here is what each one actually does, and how to decide if you need it. ## Learning mode in claude.ai Anthropic launched this as part of [Claude for Education](https://www.anthropic.com/news/introducing-claude-for-education) on April 2, 2025. It started as a Projects feature for students at partner schools like Northeastern, LSE, and Champlain College. By August 2025, Anthropic rolled it out to every claude.ai user, not just education accounts. The idea is simple. Instead of answering your question directly, Claude asks you questions back. If you ask it to solve a calculus problem, it asks what approach you would try first. If you submit an argument, it asks what evidence backs your claim. This is the Socratic method, and Anthropic built the whole style around it. You turn it on from the style selector in claude.ai. Click "Search and tools" at the bottom of the chat, go to "Use style," and pick "Learning" from the list. It stays on for that conversation until you switch it off. In my experience, this mode is genuinely more useful for open-ended reasoning than for factual lookups. Ask it to help you understand a proof, and it earns its keep. Ask it a quick factual question, and the back-and-forth just slows you down. ## Learning mode in Claude Code This is a different feature entirely, and it ships as one of [four built-in output styles](https://code.claude.com/docs/en/output-styles) in Claude Code: Default, Proactive, Explanatory, and Learning. Default is the normal system prompt you already know. Explanatory adds educational "Insights" while Claude codes, explaining why it picked an approach. Learning goes further. Claude still writes most of the code, but it drops `TODO(human)` markers at key points. Those markers ask you to write the implementation yourself, with guidance on how to approach it. I tried it on a small refactor and the pattern held up. Claude explained the surrounding architecture, then left a function stub with a `TODO(human)` comment and a hint about the approach. I filled it in, Claude reviewed it, and we moved on. It felt closer to pairing with a senior engineer who lets you drive the tricky part than to an assistant that writes everything for you. To switch to it, run `/config`, select "Output style," and choose "Learning." Output style is baked into the system prompt at session start, so the change applies only after you run `/clear` or start a new session. The older standalone `/output-style` command is gone now too, removed in Claude Code v2.1.91, so `/config` is the only path. This style is worth reaching for when you are onboarding a junior developer onto a codebase. It also helps when you are learning a new language or framework yourself and want forced practice instead of copy-pasted output. For a routine feature you just need shipped, switch back to Default. The extra insights and TODO markers cost you time and output tokens you do not always want to spend. If you have not looked at Claude Code's other built-in styles and features, [the features worth learning first](/blog/best-super-powers-for-claude-code/) covers the ones I use daily alongside output styles. ## How the two compare to ChatGPT's study mode OpenAI shipped a similar feature called Study Mode for ChatGPT. Reviewers keep landing on the same rough difference. ChatGPT's version moves fast, taking you from confused to competent, which is good for exam cramming and quick review. Claude's learning mode pushes back harder and stays coherent over long sessions. That suits the slower work of understanding something difficult, rather than just passing a test on it. Neither one solves the real problem with AI-assisted studying. A good conversation with a chatbot does not create a memory the way a written note does. If you want retention, save the useful exchanges somewhere outside the chat window. This is the same reason I keep persistent memory files for my own Claude Code setup. I cover that setup in [my harness breakdown](/blog/claude-code-harness-setup-that-works/), and it beats relying on chat history to carry context forward. ## Where this fits if you already use Claude for work If you run Claude across separate projects for research, writing, or client work, learning mode is not a separate product. It is just another style you can attach to any one of those conversations. I go through how I split up Claude Projects for different parts of a business in [my 5-project setup](/blog/5-claude-projects-run-your-business/). Learning mode slots into the research project cleanly, for whenever I want to actually understand a topic instead of skimming a summary. ## Try it on your next hard problem Pick something you are stuck on, not something you already know the answer to. Turn on Learning in claude.ai's style menu for a concept you want to understand. Switch Claude Code to the Learning output style for a piece of code you want to write yourself with backup nearby. The mode only pays off when you actually let it slow you down.