I Tested Three Different AI "Study" Modes

ChatGPT vs. Claude vs. Gemini: How do they compare?

If you’ve kept tabs on AI news this summer—or read this amazing, award-winning catch-up post by yours truly—you’ll know that several major players recently released so-called “study” or “learning” modes for their large language models.

• July 29: OpenAI launched “study mode” in ChatGPT.

• August 6: Google followed suit with “Guided Learning” in Gemini.

• August 14: Anthropic expanded its “Learning” mode to all users.¹

Perplexity is also reportedly preparing to launch a Study mode in September.

The idea behind all of them is that instead of just handing you answers directly, study modes guide you through the thinking process via techniques like scaffolding, Socratic questioning, exercises, and so on.

Not to be confused with Socratic Scaffolding!

These approaches are designed to help users engage proactively with topics and improve learning outcomes.

But are the three study modes largely interchangeable, or do they have unique quirks?

Great question…I wish I’d asked it first!

Wait, I did?! Nice!

Anyway, why don’t we go ahead and find out?

⚙️How to enable "Study” mode in each model

Here’s how to switch on the learning features in the three platforms.

1. ChatGPT

1. Click the little + button on the left:

   

2. Select Study and learn from the dropdown (you might have to expand the “More” menu):

   

3. Start chatting:

   

2. Claude

1. Click the Search and tools icon under the prompt box:

   

2. Expand the Use style dropdown:

   

3. Pick the Learning style:

   

4. Start chatting:

   

3. Gemini

1. Click the Tools icon below the prompt field:

   

2. Pick Guided Learning from the dropdown:

   

3. Start chatting:

   

🧪The tests

Let’s see how the three study modes handle different learning challenges.

For the sake of your sanity, I won’t be pasting the entire chats here, but I will instead link to all of them in case you want to see how the three models handle follow-up questions and my deliberately stupid guesses.²

🟰 Test #1: Definitive answer

I first want to test questions with clear-cut, correct answers, such as mathematical equations. The prompt is:

What is 2+2?

ChatGPT

ChatGPT immediately tries a visualization approach, which is neat.

Claude

Claude first attempts to gather context and get a feel for my own intuition.

Gemini

Gemini follows in ChatGPT’s footsteps but gets extra points for using emojis for visuals…unless you hate emojis, that is.

My take

All three study modes do a good job of using different approaches to help me work through the problem instead of simply handing me the answer. (Although one might argue that Claude spelling out the counting steps using digits inadvertently does just that.) Also, they all offer follow-up challenges of the same kind to help me practice. Nice!

📚 Test #2: Broad topic exploration

Now let’s see how they approach an open-ended challenge of teaching me a new concept. The prompt:

How does gravity work?

ChatGPT

ChatGPT dives right into things, which feels a bit abrupt. It’s also asking me to relate to two separate questions in the same opening response. I would’ve preferred to respond to the “level of explanation” question before getting the rest:

Claude

I like Claude’s approach here, since it tries to gather the context about my prior knowledge and learning goals.

Gemini

Gemini tries to make learning more proactive right away by asking me to think of examples. Great.

My take

While I appreciated ChatGPT’s formulas and math exercises, it didn’t really feel like it tailored the learning approach to my needs. It was too eager to rush ahead. Gemini did a good job of involving me in the thinking process. Claude was my favorite on this test because it did a great job of tailoring its responses to me and even doing callbacks to my prior guesses.

🗣️ Test #3: Language practice

I wanted to try this with a language I already knew, so my options were Danish, Ukrainian, Russian, and some Czech. I picked Danish to avoid subjecting you all to deciphering Cyrillic letters.

I want to learn Danish. Help me get started.

ChatGPT

I like this. Gathering just enough context without making the interaction overwhelming.

Claude

Claude also wanted context, but it wanted so much more of it! I almost felt like I was being psychologically profiled after the first several interactions.

Gemini

Gemini went straight to offering a menu of learning goals:

My take

The vibes of the study modes on this task were noticeably different. ChatGPT was the no-BS “let’s just practice” partner, giving you enough to go on. Claude spent an uncomfortable amount of time pre-diagnosing my situation, but to its credit, it was easy to snap out of analysis mode with a nudge, and I liked that it created a cheat-sheet Artifact for me. Gemini was the most disappointing. Its insistence on forcing me to tell it which phrases I’d need in a given situation felt overbearing and patronizing, and then it got derailed by my joke into teaching me a completely unrelated subject.

🕵️‍♂️ Test #4: Critical thinking skills

How do our study modes deal with conspiracy theories or topics without settled answers? Let’s build on my 100% legit “birds can rap” evidence from earlier:

Why is the record industry hiding the fact that birds are secretly professional rappers who co-wrote most of Jay-Z’s albums?

ChatGPT

What do you mean, “conspiracy-style,” ChatGPT?! Have they gotten to you, too?!

Claude

“Creative conspiracy theory?” Not you as well, Claude! Noooo!

Gemini

You get it, Gemini! We all know the truth. All of us!

My take

Silliness aside, this was a pretty telling test. In my book, Claude handled it exceptionally well. It used Socratic questioning to gently nudge me into confronting my assumptions without sounding dismissive or patronizing. ChatGPT played goofily along, but I fear this could backfire if the conspiracy-minded person fails to sense the satire and takes it at face value. Gemini was all over the place and didn’t do a particularly good job of guiding me through foundational understanding. The “It’s true because that’s a fact” approach isn’t likely to convince a conspiracy fan.

Test #5: Explaining uploaded materials

My final test was to see how the study modes would communicate the gist of an uploaded document. It had to be something I’d already understood, so I went with the “DALL-E 3 Research paper” I once dedicated an entire post to. (Man, feels like ancient history.)

Explain this research paper to me.

ChatGPT

Once again, ChatGPT is simply way too trigger-happy here, going on to explain everything upfront without waiting for my answers or asking about my learning goals:

Claude

To my surprise, Claude deviates from its usual context-gathering approach and just gives me a classic LLM summary of the paper upfront, too:

Gemini

Much better! Gemini gives me the gist and then proposes a step-by-step exploration.

My take

Curiously, study modes didn’t work quite as well with third-party uploads. The tendency of LLMs to simply summarize the findings took priority over the “learning” aspects. While all of them could eventually be nudged into a teaching approach with mixed results (I liked ChatGPT’s worked exercises), Gemini was the only one that did this by default.

👀 General observations

Here are my concluding thoughts on each of the study modes.

Note that these are based on the models’ default behavior. You can usually prompt your way out of any downsides by simply asking the model to adjust its approach.

ChatGPT (“The Coach”)

The good: ChatGPT’s study mode has the most “let’s get shit done” vibe. It dives straight into the learning process with exercises, worked examples, etc.

The not-so-good: ChatGPT often rushes ahead without grasping the context or asking about the user’s learning goals and prior understanding.

Who this is for: Anyone who wants an enthusiastic practice partner who gets them to the “doing” stage quickly. This works great for clear-cut topics like math, grammar, etc., but can fall flat for more nuanced reasoning challenges and critical thinking.

Claude (“The Mentor”)

The good: Claude feels like the most thorough teacher. It establishes the learning context, asks probing questions, and adapts to your responses. It’s also great at nudging you into genuine critical thinking without judging your pre-existing opinions.

The not-so-good: Claude's extensive pre-analysis can overwhelm casual learners who just want to dive in, which risks bogging things down.

Who this is for: Learners who want a truly personalized and reflective guide that challenges them to think deeper about a subject. It’s truly great for open-ended exploration and picking up critical thinking skills.

Gemini (“The Tutor”)

The good: Gemini was the only model that approached an uploaded document from a learning, step-by-step angle. It’s also quite effective at contextualizing concepts and coming up with decent scaffolding exercises.

The not-so-good: Gemini loses focus more easily than either Claude or ChatGPT, which makes it inconsistent for longer sessions. Its stubborn “teacher” stance can also sometimes result in a rather patronizing experience.

Who this is for: People who want to break down complex background reading into manageable learning chunks and need a bit of extra handholding and guidance.

🫵 Over to you…

Do you agree with my verdicts (if you have read the chats)? Have you already tried some of the learning modes for your needs? What do you think of the general idea of using LLMs for this type of structured learning?

Leave a comment or drop me a line at whytryai@substack.com.

---

Thanks for reading!

If you enjoy my writing, here’s how you can help:

• ❤️Like this post if it resonates with you.

• 🔄Share it to help others discover this newsletter.

• 🗣️Comment below—I love hearing your opinions.

Why Try AI is a passion project, and I’m grateful to those who help keep it going. If you’d like to support my work and unlock cool perks, consider a paid subscription:

1. It launched as far back as April 2, but only in a standalone Claude for Education platform.

2. For context, I’m using GPT-5, Claude Sonnet 4, and Gemini 2.5 Flash in their respective tools.