AI and learning

ChatGPT study mode: how to practice step by step

ChatGPT study mode uses guiding questions to help you reason. Learn to request practice, check responses, and turn reviewed concepts into flashcards.

What study mode is and what OpenAI announced

ChatGPT study mode is designed to guide you with questions, break problems into steps, and check your understanding. OpenAI launched it in July 2025 for free and paid users. You can turn it on for practice and switch it off during a conversation, but you still need to check its responses.

According to the official announcement, this behavior relies on system instructions developed with teachers and pedagogy experts. It does not guarantee that the assistant will always wait for your answer or catch every mistake. OpenAI warns that behavior can be inconsistent and information can be wrong. This guide proposes keeping your own attempt at the center of the session.

Why it answers with questions instead of the answer

A guiding question gives you a chance to state the procedure before seeing a solution. Answering can reveal where you get stuck and which condition you forgot. To use that opportunity, explain what you tried, even when you are unsure. A short answer with a specific mistake provides something concrete to review.

State your goal and course level to guide the conversation: “I am reviewing unit conversions for high school chemistry.” If the questions are too easy or too advanced, say so and request an adjustment. To check an exact fact or an assignment’s wording, return to the course source; a fluent conversation does not replace that check.

Ask it to examine you, not to solve for you

Start with the problem and your attempt, then ask for one question at a time. For example: “Here is my work. Help me identify what to review before showing a solution.” These instructions express how you want to practice; they do not guarantee that the assistant will follow them.

If you receive too much information, pause the conversation and narrow the task to one step. Answer before requesting another hint. If the tool reveals the result, attempt a new variation without looking at it to check what you can do independently.

  • “Do not give me the answer. Ask one question at a time and wait for my attempt. Point out the step where I make a mistake.”
  • “Check my procedure step by step. If a step is wrong, say which one and why, and do not correct the later steps until I try again.”
  • “Give me an example different from the textbook, then ask me to explain which rule it illustrates.”
  • “Give me three multiple-choice and three open-ended questions on what we reviewed, and hide the answer until I write mine.”

A worked example: a scientific notation slip

This is an illustrative example, not a test we ran with ChatGPT. A student needs the mass of 3.2 × 10⁻⁴ mol of a compound with a molar mass of 58 g/mol. She correctly writes mass = amount of substance × molar mass, but records the multiplication as 185.6 × 10⁻⁴ g = 1.856 × 10⁻⁴ g.

A useful question would be: “If you divide the coefficient by one hundred, how must the power of ten change to keep the value equal?” The correct conversion is 185.6 × 10⁻⁴ g = 1.856 × 10⁻² g = 0.01856 g. Dividing the coefficient by one hundred requires increasing the exponent by two. The error was in normalizing the notation, not choosing the formula.

She checks the units: mol × g/mol gives grams. Next, she estimates 3 × 10⁻⁴ × 60, approximately 0.018 g, and confirms that the result has a reasonable order of magnitude. An estimate does not replace the calculation, but it can help catch a large discrepancy.

To finish, she answers a flashcard: “When I divide the coefficient by one hundred, what exponent change preserves the value?” Then she attempts another conversion before checking its solution. The goal is to test the procedure on a different example.

The limits: it can err, and it can steer you

In Edutopia, Erik Ofgang described a practical review of study mode. He found useful help correcting a scientific notation error, alongside problems with pacing and difficulty. During a writing task, the assistant even supplied a rewritten paragraph. These are observations from his tests, not experimental evidence of improved learning.

These observations sit alongside OpenAI’s stated limitation: system instructions can produce inconsistent behavior and mistakes. If you receive praise without a sufficient check, request a different question and justify your answer. The assistant saying you understand does not establish that you can apply the concept without help.

How to verify what it produces

Check the facts against your source, redo the reasoning without looking at the answer, and keep your steps. If you obtain a number but cannot explain the procedure, record that difficulty for review. A calculation can be correct while its explanation is incomplete, so check both.

When the assistant generates questions or review material, check it the way you would check an exam paper: that each question is answerable from your source, that the figures match, and that the wording has not been quietly changed. If you cannot check it, do not turn it into study material.

  • Match every figure and unit against the material it came from.
  • Rewrite the worked steps by hand and compare, instead of reading the solution.
  • Mark any explanation you could not verify, so you can raise it with your instructor.
  • Keep your work so you can compare it with a new attempt in another session.

Pair it with flashcards and spaced repetition

After a conversation, select one concept to check again. Prepare a flashcard with a specific question and a verified answer. Keep full solutions for separate exercises when you need to practice several steps. This helps you distinguish remembering a rule from applying it to a problem.

For this example, ask how the exponent changes when you divide the coefficient by one hundred and why the value stays equal. Avoid putting the full rule on the front, where you would simply read it. Space reviews across sessions and pair cards with new problems rather than assuming one correct answer establishes lasting mastery.

Connect your practice with notes and Lumia AI

Start with a question from your notes, practice answering guiding questions, and keep the concept you verified. In Lumia AI, you can create or edit flashcards, generate AI cards and quizzes from notes or PDFs, and organize them into books. Spaced-repetition review is based on your ratings.

Always check generated content before studying it and correct ambiguous questions. Lumia AI also offers a chat tutor, learning paths, Anki .apkg deck import, and web and mobile apps. Keep your written steps and sources in your own notes so you can check them again when needed.

Sources and scope of this guide

These references provide context on studying and learning. Examples are illustrative; they are not an evaluation of Lumia AI and do not guarantee grades or retention rates.

Editorial approach: the body names the source beside evidence-based claims; this list identifies the document, link, and scope consulted. We distinguish those references from our own examples. If you find an error, include the page and sentence when you contact the team.

Keep learning

AI and learning · 5 min read

AI detectors in academic work: limits and next steps

An AI detector can flag human writing by mistake. Learn what the cited research shows and how to keep drafts and explain your academic process.

Lumia AI teamPublished:

AI and learning · 5 min read

How to cite ChatGPT in APA: a step-by-step guide

The APA reference for ChatGPT with author, date, version, descriptor, and URL, plus the in-text citation, prompt records, and frequent mistakes.

Lumia AI teamPublished:

Turn each explanation into flashcards

Generate flashcards and quizzes from your notes in Lumia AI, review the content, and revisit it based on how well you remembered.

Get started free