Tutoring
Is ChatGPT a good tutor?
It is an excellent explainer and a poor teacher, and the difference is not about model quality. A tutor decides what to teach, checks it landed, and refuses to do the work.
5 min read
Key takeaways
- ChatGPT is very good at explaining and poor at teaching. Explaining is answering a question; teaching is deciding which question needed asking.
- It waits to be asked, which fails exactly the students who do not know what they do not know.
- It cannot tell whether you understood. Your saying "got it" is the only signal it has.
- It will do your work if you ask, and the thing that produces learning is the struggle it just removed.
- Used deliberately - as an explainer you interrogate, and a quizzer you cannot bluff - it is genuinely useful.
ChatGPT is an excellent explainer and a poor teacher. It will clarify almost anything you ask about, at any hour, patiently and usually accurately. It will not decide what you need to learn, notice that you have misunderstood, refuse to do your homework, or remember in April what you got wrong in February. Those four absences are what separate explaining from teaching, and they are product decisions rather than limits of the model.
What it is genuinely good at
Credit where it is due, because the useful version of this question is not "is AI bad".
- Explaining the same thing a different way. The most valuable thing a confused student can ask for, and the thing a textbook cannot do.
- Answering the embarrassing question. Students ask an AI the thing they would not raise in class, and that has real value.
- Availability. 11pm, the night before, when no tutor exists.
- Pitching to a level. Ask for it as though you were twelve, and it obliges.
- Infinite practice problems. Within reason, and with checking.
A student who uses it only for these is better off than one who does not.
The four things it structurally does not do
It waits to be asked. This is the deep one. A chatbot is reactive by design, so the quality of your learning is bounded by the quality of your questions. The students who most need help are precisely those who cannot formulate the question, because not knowing what you do not know is what being a novice means. A teacher's first job is deciding what you should be working on, and that job is not on offer.
It cannot tell whether you understood. You say "got it", and it believes you. There is no check. And students systematically overestimate their own understanding, because material that feels easy to process feels learned - the fluency illusion that makes rereading the most popular and one of the least effective study techniques.
It will do the work. Ask for the answer and you get the answer. The difficulty you just removed was the thing producing the learning, and this is the single most damaging ordinary use.
It does not remember what matters. Even with memory features, there is no model of you as a learner: which concepts are solid, which are shaky, what is due for review. Every session starts roughly from zero.
Explaining against teaching
| ChatGPT | A teacher | |
|---|---|---|
| Decides what to study | You do | They do |
| Checks understanding | Asks if you understood | Makes you demonstrate it |
| When you are stuck | Gives the answer | Decides whether to let you struggle |
| Memory | This conversation | Months of specific evidence |
| Wrong ideas | Corrected if raised | Hunted deliberately |
| Pacing | Your attention span | A plan, with review scheduled |
Every row is a decision somebody made about what the product is for. A general assistant should answer what it is asked; that is the right design for a general assistant and the wrong one for a teacher.
How to use it well
If it is what you have, use it like this.
Make it explain, then interrogate it. Never stop at the first answer. "Why?" "What if that part changed?" "What is the most common mistake here?" The interrogation is where the learning is.
Make it quiz you, strictly. "Ask me ten questions on this one at a time. Do not give me the answer until I have attempted it." You must answer before revealing, or the exercise is reading with extra steps.
Explain it to the machine. Write out your understanding of a topic in your own words, paste it in, and ask where it is wrong, incomplete or vague. This is the one pattern that uses the model's strength while keeping the hard work yours - and self-explanation is itself one of the better-evidenced study behaviours.
Never paste in the problem set. If you cannot reproduce the work without it, you have not learned it, whatever the mark says.
Prompts that actually work
If a general assistant is your tool, these three patterns do most of the available good. They work because each one keeps the hard part with you.
The interrogation. Ask for an explanation, then refuse to accept the first one.
Explain why the bond angle in water is 104.5 degrees rather than 109.5. Now explain why that matters for the molecule being polar. What is the most common mistake students make here? Ask me a question that would catch me if I had only memorised this.
The strict quizmaster. The instructions matter, because the default behaviour is to be helpful and give you the answer.
Quiz me on cellular respiration, one question at a time. Wait for my answer before responding. Do not give me the answer or any hint until I have attempted it. After I answer, tell me what was missing and ask the next question.
The hole-finder. This one exploits the model's real strength while keeping the work yours.
Here is my understanding of meiosis, written from memory. Point out what is wrong, what is missing, and what is vague. Do not rewrite it for me.
Notice what all three have in common: you produce something first. A session where the model produces and you read is reading with extra steps.
How much to trust what it says
On well-established school and introductory university material, accuracy is usually good. The failure mode that matters is not frequency but texture: a wrong explanation arrives in exactly the same confident register as a right one, and a student learning the topic has no way to tell them apart. That is a different risk from a textbook being wrong, because a textbook is at least wrong consistently and somebody eventually notices.
Three rules that cost little:
- Verify anything numerical. Figures, dates, constants, study results.
- Be sceptical of anything specific to your course. It does not know your syllabus, your professor's emphasis, or your exam board, and it will answer confidently anyway.
- Treat a citation as a claim, not a source. Check that the paper exists and says what the model said it says.
When a general assistant is genuinely the wrong tool
- When you do not know what to study. It cannot tell you, and that is the question that matters most.
- When you are behind and panicking. It will answer whatever you ask, which at 1am is the assignment rather than the concept.
- When you have a persistent misunderstanding. It cannot see it, because it has no record of your previous attempts.
- When the work is the point. Essays, problem sets and proofs are the learning, not the deliverable.
Where TruLearn fits
TruLearn exists because of the four gaps above, and each is a deliberate inversion. It decides what to teach before the student speaks, from what they got wrong last time and what is due for review. It checks understanding by making the student explain the idea back out loud, graded against a rubric a person wrote for that concept. It is built to let a student struggle rather than rescuing them, and it will not do a problem set. And it stores the wrong idea in the student's own words, then brings it back days later.
It teaches AP Biology and introductory college biology, and nothing else yet. See how a session works, or AI tutor versus human tutor.
Frequently asked questions
- Can ChatGPT replace a tutor?
- No, though it can replace part of what a tutor does. It explains clearly and is always available. It does not decide what you should study, does not verify that you understood, has no memory of your specific weaknesses across months, and will hand you the answer if you ask. A tutor's value is concentrated in exactly those four things.
- Is it accurate enough to study from?
- Usually, on well-established material at school and introductory university level, and not reliably enough to trust unverified on anything specialised, numerical or recent. The failure mode that matters is confidence: a wrong explanation arrives in the same fluent tone as a right one, and a student learning the topic cannot tell them apart.
- Is using ChatGPT for homework cheating?
- Having it do the work is. Having it explain a method you then apply yourself is not, in most institutions' rules. The practical test is whether you could reproduce the work without it. If not, you have outsourced the learning rather than supported it.
- What is the best way to use it for studying?
- Three ways that work. Ask it to explain something you did not follow, then ask why repeatedly. Ask it to quiz you without giving the answers, and make yourself answer before revealing. And explain a topic to it in your own words and ask it to find the holes.
- How is an AI tutor different from ChatGPT?
- A purpose-built tutor decides what to teach before you say anything, follows a curriculum mapped to standards, checks understanding by making you explain rather than by asking if you understood, declines to do your work, and remembers your specific errors across sessions. Those are product decisions rather than model capabilities.
Keep reading
Tutoring
What the research says actually makes tutoring work
Tutoring is one of the best-evidenced interventions in education, but the effect depends on who tutors, how often, and when. Most purchased tutoring misses on frequency.
Oct 3, 20265 min read
Tutoring
How to choose a tutor
Most families choose on price and availability, then hope. Five questions, asked before you pay, that predict whether tutoring will work.
Oct 3, 20265 min read
Tutoring
What does a tutor cost, and what are you paying for
Federal wage data puts the median tutor at about $21 an hour, while private tutoring is commonly advertised far above that. The gap explains more than the price does.
Oct 3, 20266 min read