ChatGPT Study Mode: what it's good at, where it falls apart, and how to actually turn it on

What it actually is

OpenAI launched Study Mode on 29 July 2025, built with input from teachers and learning scientists, and the pitch has stayed consistent since: instead of answering a question directly, it asks what you already know, breaks the topic into smaller steps, and checks your understanding along the way rather than handing you a finished answer.

The mechanism is simpler than the marketing implies. Study Mode is a system prompt, not a distinct model — it runs on top of whatever model you already have selected in ChatGPT, GPT-5.6 Luna or otherwise. The actual instruction text has leaked and circulates publicly on GitHub; a Hacker News commenter noted you can get a close approximation just by asking ChatGPT to repeat its current system instructions back to you in a code block, or by pasting the same behavior into a Project's custom instructions. That's not a criticism of the feature so much as useful context for what you're actually getting: a well-crafted prompt, not a specialized tutor built from scratch.

How to turn it on, and why some guides are wrong

As of August 2026, three ways work: type @study in the composer, click the + button and search for "study" in the tools menu, or go directly to chatgpt.com/studymode. It's available on Free, Plus, Pro and Team from launch, with ChatGPT Edu access following shortly after.

If you've read a guide telling you to type /study, that was the original activation method and it's the reason a real controversy happened. Around 12 April 2026, users started reporting on Hacker News and several blogs that Study Mode had been "silently removed" — the slash command stopped surfacing in the tools menu, with no changelog entry and no announcement. It even briefly appears to have dropped from the mobile app's visible menu per at least one independent write-up. Wharton professor Ethan Mollick, a prominent voice on AI in education, posted about the apparent removal, then followed up on 11 July 2026 to clarify what had actually happened: "ChatGPT still has study mode, but rather than /study you now have to type @ study." The feature was never deleted. OpenAI just changed the trigger without telling anyone, for three months.

The practical lesson: OpenAI has already changed how you access this feature once without notice. If @study stops working by the time you're reading this, try the direct URL first before assuming it's gone.

What it's actually good at

Education writer Erik Ofgang ran the most useful hands-on test available, working through scientific notation, photosynthesis, calculus and a college-level argumentative essay for Edutopia. On the narrow, well-defined problems, Study Mode did what it promises: when he deliberately made a calculus error, it caught the mistake and walked through the correction step by step rather than just supplying the right answer. That's the use case it's built for — a single concept with a checkable right answer, where the Socratic back-and-forth actually teaches something instead of just adding friction.

• A concept with one correct answer you want to arrive at yourself — a formula, a specific calculation, a factual mechanism like how photosynthesis converts light to energy.
• Catching your own mistakes before a test, since the questioning format is well suited to surfacing exactly where your reasoning went wrong.
• A first pass at unfamiliar material where you want structure rather than a wall of text — though even here it needs managing; see below.

Where it actually falls apart

The same Edutopia test is also the most useful evidence of its real limits, because it's the one source that tried to break it honestly instead of just describing the pitch. Ofgang's summary is the sharpest available: "It's a wonderful answer machine, which could be useful to many students. However, this is exactly what it claims not to be."

• Open-ended writing tasks degrade fast. Given an argumentative essay assignment, Study Mode drifted into a list of generic negatives and formulaic fill-in-the-blank thesis templates, and at points rewrote entire paragraphs into polished prose a student could simply submit — the opposite of the guided-thinking pitch, and without needing any adversarial prompting to get there.
• Pacing is uneven. It sometimes dumps two dense, textbook-style pages at once, or assumes mastery of a step before checking whether you've actually got it, which several testers describe as needing an active "slow down" nudge rather than something it self-corrects for.
• It doesn't reliably hold onto a plan across a session. A Hacker News commenter working through a self-set syllabus described it repeatedly forgetting the plan it had proposed and never actually completing what it set out to teach. Progress tracking runs on ChatGPT's general Memory feature if you have that turned on, not a dedicated study-tracking system, so treat "it remembers where we left off" as inconsistent rather than guaranteed.
• The guardrail is soft, not enforced. Nothing stops you from switching back to a normal chat, and nothing stops the model from being talked into a direct answer — the Edutopia essay test shows this can happen organically, without anyone trying to jailbreak it.

Study Mode vs. the "Think" button — a common mix-up

These are two unrelated features that shipped close enough together to cause real confusion. Think is a reasoning-depth toggle that came with GPT-5.6 Luna, OpenAI's current default free-tier model: it makes the model spend more time and compute before answering any question, for any topic, to improve accuracy. Study Mode is a behavior layer that changes the interaction pattern itself — questions instead of answers — regardless of which model or reasoning depth is running underneath. You can use both at once: Study Mode's Socratic format, powered by a model that's thinking harder before each response.

How it compares to the competition

Every major assistant now has some version of this. None of them is uniformly better; each has a distinct failure mode, per a hands-on comparison from independent reviewer whytryai.com that tested identical prompts across all three:

Tool                Style                    Where it struggles
--------------------------------------------------------------------
ChatGPT Study Mode   Fast, direct questioning  Rushes ahead without
                                                confirming real grasp
                                                of context

Claude Learning Mode Rigorous, Socratic        Over-questions before
(opened to all users                           getting to substance;
Aug 2025)                                      can feel interrogated

Gemini Guided        Strong on document/slide  Loses focus easily,
Learning (Aug 2025)  scaffolding               can read as patronizing
                                                about exact phrasing

Khanmigo (Khan       Purpose-built tutor,      Narrower scope — built
Academy, ~$4/mo)     never gives direct        for K-12 math/STEM,
                     answers                   not general subjects

The pattern across independent testing is consistent: general chatbots retrofitted with a study mode are good at some subjects and weak at others, while a purpose-built tutor like Khanmigo is narrower but harder to trick into just giving the answer. If your subject is K-12 math specifically, that trade-off is worth knowing about.

The gap that matters for classrooms: no monitoring

There is no teacher or parent dashboard for Study Mode. Nobody can see what a student asked, what hints they got, or whether they eventually just talked the model into giving them the answer. Education Week quoted a math teacher putting the underlying accuracy risk plainly: "You'd better be pretty confident it's going to give you correct answers," a concern that doesn't go away just because the interface asks questions instead of stating facts. If you're a teacher considering this for classroom use, that combination — no visibility, plus the essay-task failure mode above — is worth weighing against a tool built for institutional oversight from the start.

When a source-grounded tool is the better fit

Study Mode is drawing on general knowledge, the same way the rest of ChatGPT does — it doesn't know your specific textbook, lecture slides or professor's rubric unless you upload them into that conversation. For learning a concept cold, that's fine. For studying material you've been assigned — a reading, a set of lecture notes, a specific paper — a tool that answers only from what you gave it and cites the passage it used is a meaningfully different, and often safer, choice: Gemini Notebook works this way, and our guide for students covers building a notebook around a course's actual material rather than the model's general knowledge of the subject. Study Mode and a grounded notebook aren't really competing for the same job — one teaches a concept, the other interrogates a specific set of sources — and using both for the same class isn't unusual.

People also ask

Is ChatGPT Study Mode free?

Yes, on every plan including Free, Plus, Pro and Team, plus ChatGPT Edu. There's no separate charge or upgrade required.

Did OpenAI remove Study Mode?

No, but it looked that way for months. Around April 2026 the original /study slash command stopped appearing in menus with no announcement, sparking a real Hacker News thread about its disappearance. It was never deleted — OpenAI switched the trigger to @study, confirmed by AI-in-education researcher Ethan Mollick in July 2026, and the direct URL chatgpt.com/studymode worked the entire time.

How is Study Mode different from GPT-5.6's Think button?

They're unrelated. Think is a reasoning-depth toggle that makes any response more thorough, for any topic. Study Mode is a Socratic interaction style — questions and hints instead of direct answers — that works with any model, including one with Think turned on.

Does Study Mode work well for essay writing?

This is its weakest documented area. A hands-on test by Edutopia found it degrading into generic negative-feedback lists and fill-in-the-blank thesis templates, and at points rewriting a student's paragraphs into finished prose — the opposite of the guided-questions pitch, without any adversarial prompting needed to trigger it.

Can a teacher see what a student did in Study Mode?

No. There's no monitoring or admin dashboard for Study Mode specifically, in ChatGPT's consumer or Edu tiers. That's a real limitation for classroom adoption, not an unlisted setting you haven't found.

Does Study Mode remember what I was studying last time?

Inconsistently. It relies on ChatGPT's general Memory feature rather than dedicated study-progress tracking, and multiple independent reports describe it losing track of an agreed study plan partway through a session. Don't rely on it to pick up exactly where you left off.

Is Study Mode better than Gemini Notebook for studying?

They do different jobs. Study Mode teaches a concept using general knowledge and a Socratic back-and-forth; Gemini Notebook answers only from documents you give it and cites the exact passage. Use Study Mode to learn something cold, and a grounded notebook when you're studying material you were actually assigned.